Marine ecological disaster early warning method based on multi-source heterogeneous big data fusion

Abstract

The invention discloses a marine ecological disaster early warning method based on multi-source heterogeneous big data fusion. The marine ecological disaster early warning method comprises the steps that remote sensing image data and actual measurement environment data of a target sea area are acquired respectively; sequentially performing boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain multispectral feature data; performing space-time alignment on the multispectral feature data and the environmental data to obtain a space-time alignment data set; mapping each sample pair data in the time-space alignment data set into graph node data, and analyzing a connection relationship between any two adjacent graph node data to construct an adjacency relationship matrix; performing neighborhood aggregation processing on the adjacency relation matrix to generate a fusion feature vector; and inputting the fusion feature vector into a pre-constructed marine disaster prediction model to obtain a multi-dimensional prediction result, and carrying out ecological disaster real-time early warning based on the multi-dimensional prediction result. According to the method, the real-time performance of early warning of the marine ecological disasters is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of marine environmental monitoring, and in particular to a method for early warning of marine ecological disasters based on the fusion of multi-source heterogeneous big data. Background Art

[0002] Marine ecological disasters seriously threaten the safety of the marine ecosystem and the sustainable development of mankind. Their occurrence is characterized by suddenness, regionality, and spatio-temporal heterogeneity, and accurate real-time early warning technology is required.

[0003] The prior art identifies the algae aggregation area through the spectral characteristics of remote sensing images. However, this method can only capture local characteristics in a single scene. Taking the red tide early warning as an example, the prior art can only judge the general distribution of algae, but cannot know the dynamic changes of key environmental factors in the water body, resulting in low accuracy of disaster prediction, and further affecting the accurate judgment of key information such as the probability of disaster occurrence and the affected range, and cannot meet the needs of real-time prevention and control of marine ecological disasters. Summary of the Invention

[0004] The present invention provides a method for early warning of marine ecological disasters based on the fusion of multi-source heterogeneous big data, so as to solve the technical problem of how to improve the existing method for early warning of marine ecological disasters, and achieve the effect of improving the real-time performance of marine ecological disaster early warning.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a method for early warning of marine ecological disasters based on the fusion of multi-source heterogeneous big data, including: Obtaining remote sensing image data and measured environmental data of a target sea area respectively; Constructing an image segmentation network including an attention mechanism and a multi-scale feature fusion module, and sequentially performing boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain optimized multi-spectral feature data; Performing spatio-temporal alignment on the multi-spectral feature data and the environmental data according to a dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set; Mapping each sample pair data in the spatio-temporal alignment data set into graph node data, and analyzing the connection relationship between any two adjacent graph node data, and constructing an adjacency relationship matrix based on the analysis result, where the connection relationship reflects the feature correlation and spatio-temporal correlation between the graph node data; Performing neighborhood aggregation processing on the adjacency relationship matrix based on a pre-constructed graph convolutional network to generate a fusion feature vector, where the fusion feature vector includes the spatio-temporal dependence relationship between the spectral feature data and the environmental data; Construct an ocean disaster prediction model according to the cross-domain transfer learning method, input the fused feature vector into the ocean disaster prediction model, obtain a multi-dimensional prediction result reflecting the early warning of ocean ecological disasters, and conduct real-time early warning of ecological disasters in the target sea area based on the multi-dimensional prediction result.

[0006] As one of the preferred solutions, the method of performing boundary segmentation and feature fusion on the remote sensing image data in sequence based on the image segmentation network to obtain optimized multi-spectral feature data includes: Extract features from the multi-spectral band data in the remote sensing image data based on the image segmentation network to obtain multi-scale spectral features, where the multi-scale spectral features include shallow spectral features and deep spectral features; Use the method of channel splicing to splice the shallow spectral features and the deep spectral features to obtain a red tide probability map with the same resolution as the remote sensing image data; Perform double-threshold segmentation on the red tide probability map to obtain red tide boundary coordinates, where the double-threshold segmentation is designed to extract the core red tide area based on the first segmentation threshold and extract the fuzzy transition boundary between the red tide and normal sea water based on the second segmentation threshold; Perform spatial association between the red tide boundary coordinates and the remote sensing image data to obtain multi-spectral feature data including multi-spectral data and the red tide boundary coordinates.

[0007] As one of the preferred solutions, the method of using the method of channel splicing to splice the shallow spectral features and the deep spectral features to obtain a red tide probability map with the same resolution as the remote sensing image data includes: Adjust the spatial dimensions of the shallow spectral features and the deep spectral features to obtain a first shallow spectral feature and a first deep spectral feature with the same resolution as the remote sensing image data; Allocate corresponding adaptive weights to the first shallow spectral feature and the first deep spectral feature respectively to obtain a second shallow spectral feature and a second deep spectral feature, where the adaptive weight reflects the importance of the shallow spectral features and the deep spectral features in red tide probability prediction; Splice the second shallow spectral feature and the second deep spectral feature in the channel dimension to obtain a red tide probability feature map; Calculate the probability that each pixel point in the red tide probability feature map is a red tide area to obtain the red tide probability map.

[0008] As one of the preferred solutions, the method of performing spatio-temporal alignment on the multi-spectral feature data and the environmental data according to the dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set includes: Taking the transit time of the remote sensing satellite as a reference, the acquisition time of the environmental data is dynamically resampled through a linear interpolation algorithm to generate first environmental data synchronized with the multi-spectral feature data; Based on the geocoding parameters of the obtained multi-spectral feature data, the acquisition location of the environmental data in the first environmental data is spatially registered, and the first environmental data is mapped to a spatial grid with the same resolution as the multi-spectral feature data to obtain second environmental data; The second environmental data is subjected to rationality verification to obtain the spatio-temporal alignment data set.

[0009] As one of the preferred solutions, measured environmental data of the target sea area is obtained, including: Based on sensors, initial environmental data of the target sea area is obtained, and an outlier detection algorithm based on density peak clustering is used to detect outliers in the initial environmental data to obtain first environmental data with data deviating from the normal distribution removed; The first environmental data is filtered through a Kalman filter algorithm to obtain smoothed second environmental data; Feature extraction is performed on the second environmental data to construct a standardized data sample including a timestamp, a spatial position, and feature parameters, where the feature parameters include a temperature change rate, a salinity gradient, a fluctuation range of dissolved oxygen content, and a pH value anomaly coefficient.

[0010] Mapping each sample pair data in the spatio-temporal alignment data set into graph node data, and analyzing the connection relationship between any two adjacent graph node data, and constructing an adjacency relationship matrix based on the analysis results, including: As one of the preferred solutions, each sample pair data in the spatio-temporal alignment data set is correspondingly mapped into graph node data in a graph structure, and each graph node data includes a spatial attribute, a time attribute, and a feature attribute; Calculating the spatial distance of the spatial attributes of any two adjacent graph node data based on the Euclidean distance formula; Calculating the feature similarity of the feature attributes of any two adjacent graph node data according to the Pearson correlation coefficient; Judging the connection relationship between any two adjacent graph node data based on the spatial distance and the feature similarity, and constructing an adjacency relationship matrix according to the connection relationship.

[0011] As one of the preferred solutions, the judging the connection relationship between any two adjacent graph node data based on the spatial distance and the feature similarity, and constructing an adjacency relationship matrix according to the connection relationship, includes: Based on the geographical environment characteristics of the target sea area, the seasonal variation law of the marine ecosystem, and the historical marine ecological disaster data, determine the spatial distance threshold and the feature similarity threshold of the target sea area; Make a preliminary judgment on the connection relationship of any two adjacent graph node data according to the spatial distance threshold and the feature similarity threshold; If the preliminary judgment result is that there is a connection relationship between the two graph node data, then calculate the connection strength between the two graph node data based on the weighted summation method to generate an adjacency relationship matrix reflecting the spatio-temporal correlation and feature correlation between the graph node data.

[0012] As one of the preferred solutions, the construction of the marine disaster prediction model according to the cross-domain transfer learning method includes: Extract the network structure and training parameters of the land disaster prediction model as the basic marine disaster prediction model, and embed a fully connected layer and a domain adversarial network module for marine ecological disaster prediction in the basic marine disaster prediction model to generate the first marine disaster prediction model; Obtain the historical data of marine ecological disasters as the target domain data, and obtain the historical data of land disasters as the source domain data. By minimizing the feature distribution difference between the source domain data and the target domain data, correct the parameters of the first marine disaster prediction model to obtain the second marine disaster prediction model; Perform data augmentation processing on the historical data of marine ecological disasters to obtain the second historical data of marine ecological disasters; Train the second marine disaster prediction model based on the second historical data of marine ecological disasters, and update the weight parameters of the second marine disaster prediction model by optimizing the loss function to obtain a trained marine disaster prediction model.

[0013] As one of the preferred solutions, the performing data augmentation processing on the historical data of marine ecological disasters to obtain the second historical data of marine ecological disasters includes: Add Gaussian noise with a mean of zero and an adaptively adjusted standard deviation to the time series feature data in the historical data of marine ecological disasters, and randomly offset the spatial position coordinate data of the historical data of marine ecological disasters to generate the first augmented data containing spatio-temporal perturbations; Construct a marine ecological data generation model based on the adversarial generative network, and generate the second augmented data conforming to the characteristics of the target sea area based on the marine ecological data generation model, where the marine ecological data generation model is designed to learn the distribution law of marine ecological data by inputting the feature vectors of real disaster samples; Perform validity verification on the first enhanced data and the second enhanced data, and merge the first enhanced data and the second enhanced data that pass the validity verification to obtain the second historical data of marine ecological disasters.

[0014] As one of the preferred solutions, the real-time early warning of ecological disasters in the target sea area based on the multi-dimensional prediction results includes: Obtain the real-time hydrological data and meteorological data of the target sea area; Input the multi-dimensional prediction results, the real-time hydrological data, and the meteorological data into a pre-constructed disaster assessment model, and determine the risk level of ecological disasters according to the output results of the model; Obtain the ecological sensitive areas of the target sea area, and determine the threatened degree of the ecological sensitive areas based on the disaster impact range in the multi-dimensional prediction results; Execute the early warning control instructions generated by the risk level and the threatened degree.

[0015] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: 1) The present invention respectively obtains remote sensing images and measured environmental data, processes the remote sensing images by using an image segmentation network with an attention mechanism and a multi-scale feature fusion module, then realizes data spatio-temporal alignment through a dynamic spatio-temporal calibration algorithm, constructs a graph node and an adjacency relationship matrix, and generates a fused feature vector through a graph convolutional network, integrating the advantages of multi-source data, taking into account the spectral features and the spatio-temporal dependence relationship of environmental data, breaking through the limitations of single data, and being able to capture the complex mechanism of the occurrence of marine ecological disasters more comprehensively and deeply, thereby improving the accuracy and reliability of the disaster prediction model.

[0016] 2) The present invention constructs a marine disaster prediction model by means of a cross-domain transfer learning method, inputs the fused feature vector into the model to obtain multi-dimensional prediction results, covering key information such as the probability of disaster occurrence and the impact range, and can perform real-time early warning of ecological disasters in the target sea area, effectively meeting the real-time prevention and control requirements of sudden marine ecological disasters. Description of the Drawings

[0017] Figure 1 It is a schematic flowchart of a method for early warning of marine ecological disasters based on multi-source heterogeneous big data fusion in one embodiment of the present invention. Detailed Embodiments

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0019] In the description of the present invention, the terms "first", "second", "third", etc. are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first", "second", "third", etc. may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise stated, the meaning of "a plurality" is two or more.

[0020] In the description of the present invention, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected, or indirectly connected through an intermediate medium, and it may be the communication inside two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are only for the purpose of illustration, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0021] In the description of the present invention, it should be noted that unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0022] An embodiment of the present invention provides a method for early warning of marine ecological disasters based on the integration of multi-source heterogeneous big data. Specifically, please refer to Figure 1 , Figure 1 which shows a schematic flow chart of the method for early warning of marine ecological disasters based on the integration of multi-source heterogeneous big data in one of the embodiments of the present invention, and it includes steps S1 - step S6: S1: Obtain the remote sensing image data and the measured environmental data of the target sea area respectively; Among them, the remote sensing image data is the surface spectral radiation information of the target sea area obtained by remote sensing platforms such as satellites and drones using electromagnetic wave sensors, which can include multi-spectral or hyper-spectral data and can reflect parameters such as sea surface temperature, algae distribution, and water turbidity. The measured environmental data is the in-situ environmental data collected in real time by sensors deployed in the target sea area, which can include water quality parameters, meteorological data, and ocean dynamic parameters.

[0023] It can be understood that a single remote sensing data can only obtain the sea surface spectral information, such as the distribution of surface algae during a red tide, and cannot know the dynamic changes of key environmental factors inside the water body; a single measured data only reflects the state of local points, such as the salinity value at a certain buoy, and it is difficult to construct a spatial correlation model for disaster occurrence. The combination of the two can form a complete data chain of "sea surface spectral characteristics - water body environmental parameters - disaster driving mechanism" to jointly support disaster analysis.

[0024] Specifically, in this embodiment, the obtained remote sensing image data is preprocessed, including: 1) Radiometric correction: Remove the influence of atmospheric scattering and sensor noise, and restore the true surface radiance; 2) Geometric correction: Based on ground control points or satellite attitude data, correct the geometric deformation of the image; 3) Band selection: Select sensitive bands according to the type of disaster. For example, for red tide early warning, pay attention to the reflectance difference between the red light (650nm) and near-infrared (850nm) bands.

[0025] Preferably, in an embodiment of the present invention, obtaining the measured environmental data of the target sea area includes: Based on the sensors, obtain the initial environmental data of the target sea area, and use the outlier recognition algorithm based on density peak clustering to detect outliers in the initial environmental data to obtain the first environmental data after removing the data deviating from the normal distribution; Perform filtering processing on the first environmental data through the Kalman filter algorithm to obtain the smoothed second environmental data; Extract features from the second environmental data, and construct a standardized data sample including timestamp, spatial position, and characteristic parameters, where the characteristic parameters include temperature change rate, salinity gradient, fluctuation range of dissolved oxygen content, and pH value anomaly coefficient.

[0026] Among them, density peak clustering is an unsupervised outlier detection algorithm based on local density and distance. By calculating the local density of each data point and the minimum distance to high-density points, outliers with significantly lower density than the neighborhood are identified. The Kalman filter algorithm is a recursive filtering method for time series data. By establishing a state space model and using the state estimate at the previous moment and the current observation value, the optimal state estimate at the current moment is recursively calculated to effectively suppress Gaussian noise.

[0027] The rate of temperature change is the change value of water temperature per unit time, reflecting the thermal dynamic stability of the marine environment; the salinity gradient is the difference in salinity within a unit spatial distance, characterizing the degree of water body mixing and the influence of runoff; the fluctuation range of dissolved oxygen content is the difference between the maximum and minimum values of dissolved oxygen concentration within a certain time window, reflecting the stability of the water body's redox environment; the pH anomaly coefficient is the degree of deviation of the current pH value from the average value in the same historical period, used to identify acid-base balance anomalies.

[0028] It should be noted that the initial environmental data collected by sensors are often affected by equipment failures, biological attachments, or environmental disturbances, with outliers and high-frequency noise. Direct use will lead to misjudgment of the model.

[0029] Specifically, in this embodiment, parameters such as water temperature, salinity, dissolved oxygen, and pH value of the target sea area are collected through sensors such as buoys, submersible buoys, and AUVs to form an initial data set containing timestamps and spatial positions.

[0030] The density peak clustering algorithm is used to set a distance threshold, where the distance threshold can be dynamically determined according to the standard deviation of historical data. Calculate the local density and distance parameters of each data point, identify and remove outliers, and obtain the first environmental data.

[0031] Apply the Kalman filter algorithm to the first environmental data, construct a state space model, and through the prediction-update recursive process, filter out high-frequency noise and retain the long-term trend to generate the smoothed second environmental data.

[0032] Calculate the rate of temperature change and the fluctuation range of dissolved oxygen according to the time window; based on the spatial positions of the sensors, use the inverse distance weighting method to calculate the salinity gradient; calculate the pH anomaly coefficient through Z-score. Integrate the timestamps, spatial positions, and the above characteristic parameters, and perform normalization processing to form a feature vector with a unified dimension.

[0033] Through three-level processing, the measured data is transformed from the original discrete points into high-quality samples containing spatio-temporal dynamic characteristics, providing a reliable data basis for multi-source data fusion and the construction of disaster prediction models, and significantly improving the ability of the early warning method to capture the complex changes in the marine environment.

[0034] S2: Construct an image segmentation network including an attention mechanism and a multi-scale feature fusion module, and sequentially perform boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain optimized multi-spectral feature data; Preferably, in an embodiment of the present invention, sequentially performing boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain optimized multi-spectral feature data includes: Extract features from the multi-spectral band data in the remote sensing image data based on the image segmentation network to obtain multi-scale spectral features, where the multi-scale spectral features include shallow spectral features and deep spectral features; Use the method of channel splicing to splice the shallow spectral features and the deep spectral features to obtain a red tide probability map with the same resolution as the remote sensing image data; Perform double-threshold segmentation on the red tide probability map to obtain red tide boundary coordinates, where the double-threshold segmentation is designed to extract the core red tide area based on the first segmentation threshold and extract the fuzzy transition boundary between the red tide and normal sea water based on the second segmentation threshold; Spatially associate the red tide boundary coordinates with the remote sensing image data to obtain multi-spectral feature data including multi-spectral data and red tide boundary coordinates.

[0035] Among them, the attention mechanism is a neural network module that simulates human visual attention. It makes the network focus on key areas through weight allocation, suppresses irrelevant information, and improves the pertinence of feature extraction. Multi-scale feature fusion includes using features at different levels of the convolutional neural network and integrating multi-scale information through splicing or weighting methods to solve the problem of inaccurate segmentation of complex boundaries by single-scale features.

[0036] It should be noted that although the multi-spectral data of remote sensing images can reflect the spectral anomalies on the sea surface, there are two major challenges. The first is the problem of blurred boundaries: the edges of red tides are affected by water mixing and illumination angles, and the spectral features show a gradual transition. Traditional single-threshold segmentation is prone to missing edges or misjudging noise; the second is the imbalance of feature scales: shallow features can locate the details of red tides, and deep features can identify the overall red tides, but when used independently, they cannot balance accuracy and integrity.

[0037] By focusing on the spectral features related to red tides through the attention mechanism, combining multi-scale feature fusion to retain detail and semantic information, and then distinguishing the core area and the transition boundary through double-threshold segmentation, the red tide boundary coordinates can be accurately extracted and associated with the original spectral data, providing high-precision spatial features for subsequent spatio-temporal alignment (matching with measured environmental data) and disaster scope deduction, and solving the problems of rough boundary segmentation and insufficient feature utilization in traditional methods.

[0038] Specifically, in this embodiment, an image segmentation network based on U-Net or DeepLab is constructed, and a spatial attention module and multi-scale convolution are embedded. Among them, Shallow feature extraction: Extract the shallow features of the multi-spectral image through a 3×3 convolutional layer, and the resolution is the same as that of the original image; Deep feature extraction: Extract semantic features through downsampling and deep convolution, with reduced resolution but enhanced semantic information.

[0039] Preferably, in an embodiment of the present invention, the shallow spectral features and deep spectral features are spliced by using the method of channel splicing to obtain a red tide probability map with the same resolution as the remote sensing image data, including: Adjust the spatial dimensions of the shallow spectral features and deep spectral features to obtain the first shallow spectral features and the first deep spectral features with the same resolution as the remote sensing image data; Respectively assign corresponding adaptive weights to the first shallow spectral features and the first deep spectral features to obtain the second shallow spectral features and the second deep spectral features, where the adaptive weights reflect the importance of the shallow spectral features and the deep spectral features in the red tide probability prediction; Splice the second shallow spectral features and the second deep spectral features in the channel dimension to obtain a red tide probability feature map; Calculate the probability that each pixel point in the red tide probability feature map is a red tide area to obtain a red tide probability map.

[0040] It should be noted that the shallow spectral features and deep spectral features of the remote sensing image are complementary. If the unweighted features are directly spliced, it may lead to the imbalance between details and semantic information. By adjusting the spatial dimension to ensure the alignment of the feature space, the adaptive weight assignment highlights the key role of different features in the red tide prediction. Finally, by fusing multi-scale information through channel splicing, a high-precision red tide probability map can be generated, solving the problem of insufficient adaptability of traditional fixed-weight fusion to complex spectral features.

[0041] Specifically, in this embodiment, after the shallow and deep feature extraction, a channel attention module is introduced to dynamically adjust the weights of the features of each band according to the red tide spectral features, and suppress the interference of irrelevant targets such as clouds and ships.

[0042] Restore the deep features to the original image resolution through upsampling, which is consistent with the dimension of the shallow features; splice the shallow and deep features along the channel dimension to generate a red tide probability feature map containing details and semantic information; through the convolutional layer and the Sigmoid activation function, calculate the probability that each pixel in the probability feature map belongs to the red tide area to obtain a red tide probability map with the same resolution. For example, a pixel value of 0.7 indicates a 70% probability of being a red tide pixel.

[0043] Next, set the thresholds. Among them, the first threshold, i.e., the high threshold, is used to extract the core red-tide area; the second threshold, i.e., the low threshold, is used to extract the transition boundary outside the core area.

[0044] For the binary image after double-threshold segmentation, use the OpenCV contour detection algorithm to extract the polygon vector coordinates, and record the minimum circumscribed rectangle of the core area and the boundary vertex coordinates of the transition area.

[0045] Align the red-tide boundary coordinates with the geocoding of the original remote sensing image to ensure that each boundary point corresponds to the multi-spectral pixels in the image; and add boundary labels to the original multi-spectral data to form multi-spectral feature data containing spectral features and boundary attributes.

[0046] S3: Align the multi-spectral feature data and the environmental data in space and time according to the dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set; Preferably, in an embodiment of the present invention, aligning the multi-spectral feature data and the environmental data in space and time according to the dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set includes: Taking the transit time of the remote sensing satellite as a reference, dynamically resample the acquisition time of the environmental data through the linear interpolation algorithm to generate the first environmental data synchronized with the multi-spectral feature data in time; Based on the geocoding parameters of the obtained multi-spectral feature data, perform spatial registration on the acquisition location of the environmental data in the first environmental data, and map the environmental data to the spatial grid with the same resolution as the multi-spectral feature data to obtain the second environmental data; Verify the rationality of the second environmental data to obtain a spatio-temporal alignment data set.

[0047] Among them, the dynamic spatio-temporal calibration algorithm aims to adjust data from different sources and with different spatio-temporal resolutions to make them consistent in the time and space dimensions, so as to achieve the effective fusion of data. The linear interpolation algorithm is a data processing method that estimates the data values at unknown time points through the linear relationship between known data points to achieve data synchronization in the time dimension. Geocoding parameters are parameters used to identify geographical spatial positions, such as longitude and latitude, projection coordinates, etc., which can determine the specific position of the multi-spectral feature data in the geographical space.

[0048] It should be noted that the multi-spectral feature data and environmental data are respectively sourced from remote sensing images and sensor measurements. There are differences between the two in terms of time and space. The multi-spectral feature data is usually obtained at the moment when the satellite passes by, with specific temporal and spatial resolutions; while the environmental data is continuously collected by sensors, and its temporal and spatial distributions are relatively discrete. Without spatio-temporal alignment, the two types of data cannot be effectively combined, which will affect the subsequent analysis and prediction of marine ecological disasters. Through spatio-temporal alignment, data from different sources can be integrated into the same spatio-temporal framework, providing an accurate and consistent data basis for subsequent model construction and disaster warning.

[0049] The spatio-temporal calibration method and accuracy index of the present invention are shown in Table 1.

[0050] Table 1

[0051] Specifically, in this embodiment, based on the passing moment of the remote sensing satellite, the target time points that need to be time-synchronized are determined. For each parameter in the environmental data, according to its collection time and corresponding value, the data value at the target time point is estimated using the linear interpolation algorithm. After linear interpolation, the first environmental data synchronized with the multi-spectral feature data in time is generated, ensuring the consistency of the two types of data in time.

[0052] The geocoding parameters of the multi-spectral feature data are obtained to determine its spatial reference system and resolution. For each collection location in the first environmental data, it is mapped into a spatial grid with the same resolution as the multi-spectral feature data according to the geocoding parameters. Methods such as nearest neighbor interpolation and bilinear interpolation can be used for spatial interpolation to convert discrete environmental data points into continuous spatially distributed data. Through spatial registration, the second environmental data is obtained, making the environmental data spatially aligned with the multi-spectral feature data, facilitating subsequent data fusion and analysis.

[0053] A rationality check is performed on the second environmental data, such as checking whether the value range of the data is within a reasonable interval and whether the change trend of the data conforms to physical laws. Some thresholds and rules can be set to judge the rationality of the data, and the data that does not meet the requirements is corrected or excluded. For example, if the temperature data shows abnormal high or low values, it may be caused by sensor failure or data transmission error, and further processing is required. After rationality verification, the final spatio-temporally aligned data set is obtained. This data set is consistent with the multi-spectral feature data in both time and space, and the data quality is guaranteed.

[0054] S4: Map each sample pair data in the spatio-temporal alignment dataset into graph node data, analyze the connection relationship between any two adjacent graph node data, and construct an adjacency relationship matrix based on the analysis results, where the connection relationship reflects the feature correlation and spatio-temporal correlation between the graph node data; Preferably, in an embodiment of the present invention, mapping each sample pair data in the spatio-temporal alignment dataset into graph node data, analyzing the connection relationship between any two adjacent graph node data, and constructing an adjacency relationship matrix based on the analysis results includes: Map each sample pair data in the spatio-temporal alignment dataset correspondingly into graph node data in the graph structure, and each graph node data includes spatial attributes, temporal attributes, and feature attributes; Calculate the spatial distance of the spatial attributes of any two adjacent graph node data based on the Euclidean distance formula; Calculate the feature similarity of the feature attributes of any two adjacent graph node data according to the Pearson correlation coefficient; Judge the connection relationship between any two adjacent graph node data based on the spatial distance and feature similarity, and construct an adjacency relationship matrix according to the connection relationship.

[0055] Among them, in the graph structure, each graph node represents a sample pair data in the spatio-temporal alignment dataset, and includes spatial attributes, temporal attributes, and feature attributes. The adjacency relationship matrix is a matrix describing the connection relationship between the nodes in the graph, and the elements in the matrix indicate whether there is a connection between two nodes and the strength of the connection.

[0056] The Pearson correlation coefficient is a statistic for measuring the linear correlation degree between two variables, and is used in this step to calculate the feature similarity between the feature attributes of adjacent graph node data. The Euclidean distance formula is used to calculate the distance between two points in space, and is used in this step to calculate the spatial distance between the spatial attributes of adjacent graph node data.

[0057] It should be noted that although the spatio-temporal alignment dataset is unified in time and space, the correlation relationship between the data is not clear. Mapping the data into graph node data and constructing an adjacency relationship matrix can represent the spatio-temporal correlation and feature correlation between the data in the form of a graph structure, which is convenient for subsequent use of graph analysis methods such as graph convolutional networks to mine potential patterns and rules in the data. The graph structure can better capture the local and global information between the data, which helps to improve the accuracy and reliability of marine ecological disaster prediction.

[0058] Traverse each sample pair data in the spatio-temporal alignment dataset, and map it correspondingly into graph node data in the graph structure. Assign spatial attributes, temporal attributes, and feature attributes to each graph node data.

[0059] Calculate the spatial distance between the spatial attributes of any two adjacent graph node data using the Euclidean distance formula, and calculate the feature similarity between the feature attributes of any two adjacent graph node data according to the Pearson correlation coefficient.

[0060] Preferably, in an embodiment of the present invention, based on the spatial distance and feature similarity, determine the connection relationship between any two adjacent graph node data, and construct an adjacency matrix according to the connection relationship, including: Based on the geographical environmental characteristics of the target sea area, the seasonal change law of the marine ecosystem, and the historical marine ecological disaster data, determine the spatial distance threshold and feature similarity threshold of the target sea area; Make a preliminary judgment on the connection relationship of any two adjacent graph node data according to the spatial distance threshold and feature similarity threshold; If the preliminary judgment result is that there is a connection relationship between the two graph node data, then calculate the connection strength between the two graph node data based on the weighted summation method to generate an adjacency matrix reflecting the spatio-temporal correlation and feature correlation between the graph node data.

[0061] Specifically, in this embodiment, based on the geographical environmental characteristics of the target sea area, the seasonal change law of the marine ecosystem, and the historical marine ecological disaster data, determine the spatial distance threshold and feature similarity threshold of the target sea area.

[0062] For any two adjacent graph node data, compare their spatial distance with the spatial distance threshold, and their feature similarity with the feature similarity threshold. If the spatial distance is less than or equal to the spatial distance threshold and the feature similarity is greater than or equal to the feature similarity threshold, it is preliminarily determined that there is a connection relationship between the two graph node data.

[0063] If the preliminary judgment result is that there is a connection relationship between the two graph node data, then calculate the connection strength between the two graph node data based on the weighted summation method. For example, the weights of the spatial distance and feature similarity can be set as and , where . Then the connection strength is expressed as:

[0064] Wherein, is the connection strength, is the spatial distance, is the spatial distance threshold, is the feature similarity.

[0065] Construct an adjacency matrix , and the element of the matrix represents the graph node and If there is no connection between two graph nodes, then ;otherwise, .

[0066] S5: Based on the pre-built graph convolutional network, the adjacency relationship matrix is ​​subjected to neighborhood aggregation processing to generate a fused feature vector, which includes the spatiotemporal dependency between the spectral feature data and the environmental data; Among them, the graph convolutional network (GCN) is a neural network specially designed for processing graph structure data. It extracts feature information from graph data by aggregating and transforming the features of nodes and their neighboring nodes in the graph. Neighborhood aggregation processing is the core operation step of GCN, which specifically involves weighted aggregation of the features of nodes and their neighboring nodes to capture the relationship and feature information between nodes.

[0067] It should be noted that after the multi-source spectral feature data and environmental data are aligned in time and space and constructed into a graph structure, the spatiotemporal correlation and feature correlation between the data have been represented by the adjacency matrix. However, the features of these data on different nodes are scattered and do not fully reflect the mutual influence between them. With the help of the neighborhood aggregation processing of the graph convolutional network, the features of the node and its neighboring nodes can be fused to mine the potential spatiotemporal dependency between the spectral feature data and the environmental data. The fused feature vector generated in this way can more comprehensively and accurately reflect the state of the marine ecosystem and provide more valuable information for the prediction of subsequent marine ecological disasters.

[0068] Specifically, in this embodiment, the pre-constructed graph convolution network usually includes multiple graph convolution layers, each layer has a specific number of neurons. Common graph convolution layers use specific convolution operations, such as Chebyshev convolution or first-order approximate convolution, to randomly initialize the weight parameters in the graph convolution network, and these parameters will be continuously adjusted and optimized in the subsequent training process.

[0069] The previously constructed adjacency relationship matrix is ​​input into the graph convolutional network as the topological structure information of the graph. The data of each graph node constitutes the node feature matrix, which also serves as the input of the graph convolutional network.

[0070] In each layer of the graph convolutional network, for each node, its neighbor nodes are determined according to the adjacency matrix. Then the features of the neighbor nodes and the features of the node itself are weighted and summed according to certain rules. Taking the first-order approximate convolution as an example, Layer Node Features The update formula is:

[0071] in, For Node Set of neighbor nodes of and are the degrees of nodes and respectively, is the learnable weight matrix of the th layer, and

[0072] is the activation function.

[0073] By stacking multiple graph convolutional layers and continuously performing neighborhood aggregation operations, information can be more widely spread in the graph. The output of each layer serves as the input of the next layer. After multiple layers of processing, the features of the nodes can incorporate information from nodes at farther distances, thereby capturing more complex spatio-temporal dependencies.

[0074] S6: Construct an ocean disaster prediction model according to the cross-domain transfer learning method, input the fused feature vector into the ocean disaster prediction model, obtain multi-dimensional prediction results reflecting ocean ecological disaster warnings, and perform real-time warnings of ecological disasters in the target sea area based on the multi-dimensional prediction results.

[0075] Preferably, in an embodiment of the present invention, constructing an ocean disaster prediction model according to the cross-domain transfer learning method includes: Extract the network structure and training parameters of the land disaster prediction model as the basic ocean disaster prediction model, and embed a fully connected layer and a domain adversarial network module for ocean ecological disaster prediction in the basic ocean disaster prediction model to generate the first ocean disaster prediction model; Obtain historical ocean ecological disaster data as target domain data, and obtain historical land disaster data as source domain data. By minimizing the feature distribution difference between the source domain data and the target domain data, correct the parameters of the first ocean disaster prediction model to obtain the second ocean disaster prediction model; Perform data augmentation processing on the historical ocean ecological disaster data to obtain the second historical ocean ecological disaster data; Train the second ocean disaster prediction model based on the second historical ocean ecological disaster data, and update the weight parameters of the second ocean disaster prediction model by optimizing the loss function to obtain the trained ocean disaster prediction model.

[0076] Among them, cross-domain transfer learning means transferring the knowledge and models learned in one domain (source domain) to another different but related domain (target domain) to improve the performance of the target domain model, especially applicable to the situation where the target domain data is scarce. The land disaster prediction model is a machine learning or deep learning model for predicting land disasters, with a certain network structure and trained parameters.

[0077] The fully connected layer is a layer in the neural network. Each neuron in the fully connected layer is connected to all neurons in the previous layer, and is responsible for performing non-linear transformation and combination on features, and is commonly used in classification or regression tasks. The domain adversarial network module is used to enable the model to learn the common features of the source domain and target domain data through adversarial training, while minimizing the feature distribution difference between them, thereby improving the generalization ability of the model in the target domain. Data augmentation processing includes performing various transformations and augmentations on the original data, increasing the diversity and quantity of the data, and improving the generalization ability and robustness of the model.

[0078] Marine ecological disaster data is usually scarce. Directly constructing and training a marine disaster prediction model may face the problem of overfitting caused by insufficient data, and the generalization ability of the model is poor. While land disaster data is relatively abundant, and there are certain similarities between land and marine ecosystems in some aspects, such as the occurrence of disasters is affected by environmental factors. Therefore, by using the cross-domain transfer learning method and leveraging the knowledge and experience of the land disaster prediction model, it can provide a better initial state for the marine disaster prediction model. At the same time, through data augmentation processing, the historical data of marine ecological disasters can be further expanded, and the generalization ability of the model can be improved, thereby constructing a more accurate and reliable marine disaster prediction model.

[0079] Specifically, in this embodiment, the network structure and training parameters are extracted from the already trained land disaster prediction model as the basic marine disaster prediction model. A fully connected layer for marine ecological disaster prediction and a domain adversarial network module are embedded in the basic marine disaster prediction model. The fully connected layer can perform further non-linear transformation and combination on the input features to adapt to the task of marine ecological disaster prediction. The domain adversarial network module is used to handle the feature distribution difference between the source domain and the target domain.

[0080] Collect historical data of marine ecological disasters as target domain data, and at the same time obtain historical data of land disasters as source domain data. Use the domain adversarial network module to minimize the difference in feature distributions between the source domain data and the target domain data through adversarial training. Specifically, the domain adversarial network module includes a feature extractor and a discriminator. The feature extractor attempts to extract the common features of the source domain and target domain data, while the discriminator attempts to distinguish whether the data comes from the source domain or the target domain. Through continuous adversarial training, the feature extractor will gradually learn a feature representation that can eliminate domain differences, thereby correcting the parameters of the first marine disaster prediction model to obtain the second marine disaster prediction model.

[0081] Preferably, in an embodiment of the present invention, the historical data of marine ecological disasters is subjected to data augmentation processing to obtain second historical data of marine ecological disasters, including: Add Gaussian noise with a mean of zero and an adaptively adjusted standard deviation to the time series feature data in the historical data of marine ecological disasters, and randomly offset the spatial position coordinate data of the historical data of marine ecological disasters to generate first augmented data containing spatio-temporal perturbations; Construct a marine ecological data generation model based on the generative adversarial network, and generate second augmented data that conforms to the characteristics of the target sea area based on the marine ecological data generation model, where the marine ecological data generation model is designed to learn the distribution law of marine ecological data by inputting the feature vectors of real disaster samples; Verify the effectiveness of the first augmented data and the second augmented data, and merge the first augmented data and the second augmented data that pass the effectiveness verification to obtain the second historical data of marine ecological disasters.

[0082] Among them, the Gaussian noise is random noise with a Gaussian distribution, its mean is zero, and the standard deviation determines the intensity of the noise. Adding Gaussian noise can simulate measurement errors and uncertainties in actual data.

[0083] It should be noted that the historical data of marine ecological disasters is usually limited in quantity, which may lead to overfitting of the model trained based on these data and poor generalization ability. Data augmentation processing can increase the diversity and quantity of data, enabling the model to learn more different feature patterns, thereby improving the generalization ability and robustness of the model. At the same time, the augmented data generated in different ways can simulate the actual situation of marine ecological disasters from multiple perspectives, further enhancing the model's adaptability to complex situations.

[0084] Specifically, in this embodiment, for the time-series feature data in the historical data of marine ecological disasters, such as the water temperature series at different times in a certain sea area, Gaussian noise is added to each data point. The mean of the Gaussian noise is zero, and the standard deviation is adaptively adjusted according to the characteristics of the data. For the spatial location coordinate data in the historical data of marine ecological disasters, such as longitude and latitude, random offsets are performed on them. Through the above operations, the first enhanced data containing spatio-temporal perturbations is generated.

[0085] Based on the generative adversarial network, a marine ecological data generation model is constructed. The generator receives a random noise vector as input and attempts to generate marine ecological disaster data that conforms to the characteristics of the target sea area. The discriminator receives the feature vectors of real disaster samples and the generated marine ecological disaster data as input and attempts to distinguish whether they are real data or generated data.

[0086] Input the feature vectors of real disaster samples into the model, and through adversarial training, enable the generator to learn the distribution law of marine ecological data. Specifically, during the training process, the parameters of the generator and the discriminator are alternately updated, so that the discrimination ability of the discriminator is continuously improved, and at the same time, the data generated by the generator is getting closer and closer to the real data distribution. After training the generation model, by inputting a random noise vector into the generator, the second enhanced data that conforms to the characteristics of the target sea area is generated.

[0087] For the first enhanced data and the second enhanced data, validity verification is carried out. For example, check whether the value range of the data is reasonable, such as whether the water temperature data is within the reasonable range of marine water temperature; check whether the physical relationship of the data conforms to the actual situation, such as whether the relationship between salinity and density is reasonable.

[0088] Some thresholds and rules can be set to judge the validity of the data, and the data that does not meet the requirements is excluded or corrected. The first enhanced data and the second enhanced data that pass the validity verification are merged with the original historical data of marine ecological disasters to obtain the second historical data of marine ecological disasters.

[0089] Input the second historical data of marine ecological disasters into the second marine disaster prediction model for training. Define a suitable loss function, such as the cross-entropy loss function or the mean squared error loss function, to measure the difference between the model prediction result and the real label. Use an optimization algorithm to optimize the loss function, and continuously update the weight parameters of the second marine disaster prediction model until the performance of the model reaches a satisfactory level, and a trained marine disaster prediction model is obtained.

[0090] Input the fusion feature vector into the marine disaster prediction model to obtain a multi-dimensional prediction result reflecting the early warning of marine ecological disasters, where the multi-dimensional prediction result includes: Disaster occurrence probability: For a specific sea area and time interval, predict the possibility of the occurrence of marine ecological disasters; Disaster type: Output the probability distribution of the disaster type; Influence scope and intensity: Output the spatial scope and severity of the disaster impact.

[0091] Preferably, in an embodiment of the present invention, real-time early warning of ecological disasters in the target sea area is performed based on multi-dimensional prediction results, including: Obtain the real-time hydrological data and meteorological data of the target sea area; Input the multi-dimensional prediction results, real-time hydrological data, and meteorological data into a pre-constructed disaster assessment model, and determine the risk level of the ecological disaster according to the output results of the model; Obtain the ecologically sensitive areas in the target sea area, and determine the threatened degree of the ecologically sensitive areas based on the disaster influence scope in the multi-dimensional prediction results; Execute the warning control instructions generated from the risk level and the threatened degree.

[0092] Among them, the real-time hydrological data are dynamic data related to water such as the water flow velocity, water level, water temperature, salinity, etc. at the current moment in the target sea area, which can reflect the current water conditions of the sea area. The real-time meteorological data are the current meteorological conditions in the target sea area, such as wind speed, wind direction, air temperature, precipitation, etc. Meteorological factors will have an important impact on the marine ecosystem. The disaster assessment model is a pre-constructed model for comprehensively analyzing multi-dimensional prediction results, real-time hydrological data, and meteorological data to evaluate the risk level of ecological disasters.

[0093] The risk level represents the classification of the possible harm degree caused by ecological disasters, such as low risk, medium risk, high risk, etc., so as to take different countermeasures. The ecologically sensitive area is an area in the target sea area where the ecosystem is relatively fragile and sensitive to disasters, such as marine protected areas, fishery aquaculture areas, etc. The threatened degree represents the degree to which the ecologically sensitive area may be affected by disasters evaluated according to the disaster influence scope in the multi-dimensional prediction results. The warning control instructions are a series of countermeasure instructions formulated according to the risk level and the threatened degree, such as issuing warning information at different levels, starting emergency responses, etc.

[0094] It should be noted that although the multi-dimensional prediction results provide preliminary information on marine ecological disasters, the marine environment is dynamically changing, and the real-time hydrological data and meteorological data can reflect the actual situation of the current sea area. By inputting these data into the disaster assessment model, the risk level of ecological disasters can be determined more accurately. At the same time, considering the particularity of ecologically sensitive areas, evaluating their threatened degree helps to take targeted protection measures. Finally, executing the warning control instructions can respond to disasters in a timely manner and reduce the impact of disasters on the marine ecosystem and human activities.

[0095] Specifically, in this embodiment, hydrological monitoring devices such as buoys and underwater sensors deployed in the target sea area are used to collect data such as water flow velocity, water level, water temperature, and salinity in real time. These devices can regularly transmit the data to the data center to ensure the timeliness and accuracy of the data. Real-time meteorological data of the target sea area, including information such as wind speed, wind direction, air temperature, and precipitation, are obtained from the meteorological department. Meteorological monitoring stations can also be set up around the sea area for independent data collection.

[0096] The multi-dimensional prediction results, real-time hydrological data, and meteorological data are input into a pre-constructed disaster assessment model. Among them, the disaster assessment model can be constructed based on machine learning algorithms. The model conducts comprehensive analysis according to the input data, considers the interactions between various factors, and outputs the risk level of ecological disasters. For example, high water temperature, low wind speed, and high nutrient content may increase the risk of red tide occurrence.

[0097] Through technologies such as Geographic Information System (GIS), information such as the location, scope, and ecological characteristics of the ecological sensitive areas in the target sea area is obtained. This information can include the boundaries of marine protected areas, the distribution of fishery farming areas, etc. According to the disaster impact scope in the multi-dimensional prediction results, an overlay analysis is performed with the location of the ecological sensitive areas. It is judged whether the ecological sensitive areas are within the disaster impact scope and the degree of influence. According to factors such as the overlapping degree between the disaster impact scope and the ecological sensitive areas, and the vulnerability of the ecological sensitive areas, the threatened degree of the ecological sensitive areas is evaluated. The threatened degree can be divided into different levels such as mild, moderate, and severe.

[0098] Comprehensively considering the multi-dimensional prediction results, real-time hydrological data, and meteorological data can more accurately evaluate the risk level of ecological disasters, reduce the situations of misjudgment and missed judgment. By evaluating the threatened degree of ecological sensitive areas, targeted protection measures can be taken to reduce the impact of disasters on ecological sensitive areas and protect the marine ecological environment. Executing the early warning control instructions can promptly activate the emergency response mechanism, take effective countermeasures, and reduce the losses of disasters to the marine ecology and human activities.

[0099] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following: 1) In the present invention, by separately obtaining remote sensing images and measured environmental data, using an image segmentation network with an attention mechanism and a multi-scale feature fusion module to process the remote sensing images, and then realizing data spatio-temporal alignment through a dynamic spatio-temporal calibration algorithm, constructing a graph node and an adjacency relationship matrix, and generating a fused feature vector through a graph convolutional network, the advantages of multi-source data are fused, the spectral features and the spatio-temporal dependence relationship of environmental data are taken into consideration, the limitation of single data is broken through, the complex mechanism of the occurrence of marine ecological disasters can be captured more comprehensively and deeply, and thus the accuracy and reliability of the disaster prediction model are improved.

[0100] 2) The present invention constructs a marine disaster prediction model by means of cross-domain transfer learning method, inputs the fused feature vectors into the model to obtain multi-dimensional prediction results, covering key information such as the probability of disaster occurrence and the affected range, and can conduct real-time early warning of ecological disasters in the target sea area, effectively meeting the real-time prevention and control requirements of sudden marine ecological disasters.

[0101] The above embodiments only express several implementation manners of the present invention, and the description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the appended claims.

Claims

1. An ocean ecological disaster early warning method based on multi-source heterogeneous big data fusion, characterized in that Including: Obtaining remote sensing image data and measured environmental data of the target sea area respectively; Constructing an image segmentation network including an attention mechanism and a multi-scale feature fusion module, and sequentially performing boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain optimized multi-spectral feature data; Performing spatio-temporal alignment on the multi-spectral feature data and the environmental data according to a dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set; Mapping each sample pair data in the spatio-temporal alignment data set into graph node data, analyzing the connection relationship between any two adjacent graph node data, and constructing an adjacency relationship matrix based on the analysis result, where the connection relationship reflects the feature correlation and spatio-temporal correlation between the graph node data; Performing neighborhood aggregation processing on the adjacency relationship matrix based on a pre-constructed graph convolutional network to generate a fused feature vector, where the fused feature vector includes the spatio-temporal dependence relationship between the spectral feature data and the environmental data; Constructing a marine disaster prediction model according to a cross-domain transfer learning method, inputting the fused feature vector into the marine disaster prediction model to obtain a multi-dimensional prediction result reflecting the early warning of marine ecological disasters, and performing real-time early warning of ecological disasters in the target sea area based on the multi-dimensional prediction result.

2. The method for marine ecological disaster early warning based on multi-source heterogeneous big data fusion according to claim 1, wherein The sequentially performing boundary segmentation and feature fusion on the remote sensing image data based on the image segmentation network to obtain optimized multi-spectral feature data includes: Extracting features from the multi-spectral band data in the remote sensing image data based on the image segmentation network to obtain multi-scale spectral features, where the multi-scale spectral features include shallow-layer spectral features and deep-layer spectral features; Using the method of channel splicing to splice the shallow-layer spectral features and the deep-layer spectral features to obtain a red tide probability map with the same resolution as the remote sensing image data; Performing double-threshold segmentation on the red tide probability map to obtain red tide boundary coordinates, where the double-threshold segmentation is designed to extract the core red tide area based on a first segmentation threshold and extract the fuzzy transition boundary between the red tide and normal sea water based on a second segmentation threshold; Performing spatial association on the red tide boundary coordinates and the remote sensing image data to obtain multi-spectral feature data including multi-spectral data and the red tide boundary coordinates.

3. The method for early warning of marine ecological disasters based on multi-source heterogeneous big data fusion according to claim 2, wherein The using the method of channel splicing to splice the shallow-layer spectral features and the deep-layer spectral features to obtain a red tide probability map with the same resolution as the remote sensing image data includes: Adjusting the spatial dimensions of the shallow-layer spectral features and the deep-layer spectral features to obtain a first shallow-layer spectral feature and a first deep-layer spectral feature with the same resolution as the remote sensing image data; Respectively assigning corresponding adaptive weights to the first shallow-layer spectral feature and the first deep-layer spectral feature to obtain a second shallow-layer spectral feature and a second deep-layer spectral feature, where the adaptive weight reflects the importance degree of the shallow-layer spectral features and the deep-layer spectral features in red tide probability prediction; Splicing the second shallow-layer spectral feature and the second deep-layer spectral feature in the channel dimension to obtain a red tide probability feature map; Calculate the probability that each pixel point in the red tide probability feature map is a red tide area to obtain the red tide probability map.

4. The method for marine ecological disaster early warning based on multi-source heterogeneous big data fusion according to claim 1, wherein, The spatio-temporal alignment of the multi-spectral feature data and the environmental data according to the dynamic spatio-temporal calibration algorithm to obtain a spatio-temporal alignment data set includes: Based on the transit time of the remote sensing satellite, the acquisition time of the environmental data is dynamically resampled by a linear interpolation algorithm to generate first environmental data synchronized with the multi-spectral feature data. Based on the geocoding parameters of the obtained multi-spectral feature data, the acquisition location of the environmental data in the first environmental data is spatially registered, and the first environmental data is mapped to a spatial grid with the same resolution as the multi-spectral feature data to obtain second environmental data. Verify the rationality of the second environmental data to obtain the spatio-temporal alignment data set.

5. The method for marine ecological disaster early warning based on multi-source heterogeneous big data fusion according to claim 1, wherein, Obtain the measured environmental data of the target sea area, including: Based on the sensors, obtain the initial environmental data of the target sea area, and use an outlier recognition algorithm based on density peak clustering to detect outliers in the initial environmental data to obtain first environmental data excluding data deviating from the normal distribution. Perform filtering processing on the first environmental data through a Kalman filter algorithm to obtain smoothed second environmental data. Extract features from the second environmental data and construct a standardized data sample including a timestamp, a spatial position, and feature parameters, where the feature parameters include a temperature change rate, a salinity gradient, a dissolved oxygen content fluctuation amplitude, and a pH value anomaly coefficient.

6. The method for marine ecological disaster early warning based on multi-source heterogeneous big data fusion according to claim 1, wherein, Mapping each sample pair data in the spatio-temporal alignment data set into graph node data, and analyzing the connection relationship between any two adjacent graph node data, and constructing an adjacency relationship matrix based on the analysis result, including: Correspondingly map each sample pair data in the spatio-temporal alignment data set into graph node data in a graph structure, and each graph node data includes a spatial attribute, a time attribute, and a feature attribute. Calculate the spatial distance of the spatial attributes of any two adjacent graph node data based on the Euclidean distance formula. Calculate the feature similarity of the feature attributes of any two adjacent graph node data according to the Pearson correlation coefficient. Judge the connection relationship between any two adjacent graph node data based on the spatial distance and the feature similarity, and construct an adjacency relationship matrix according to the connection relationship.

7. The marine ecological disaster early warning method based on multi-source heterogeneous big data fusion according to claim 6, characterized in that, The judgment of the connection relationship between any two adjacent graph node data based on the spatial distance and the feature similarity, and the construction of an adjacency relationship matrix according to the connection relationship, includes: Based on the geographical environment characteristics of the target sea area, the seasonal change law of the marine ecosystem, and historical marine ecological disaster data, determine the spatial distance threshold and the feature similarity threshold of the target sea area. Make a preliminary judgment on the connection relationship of any two adjacent graph node data according to the spatial distance threshold and the feature similarity threshold. If the preliminary judgment result is that there is a connection relationship between the two graph node data, then calculate the connection strength between the two graph node data based on the weighted summation method to generate an adjacency relationship matrix reflecting the spatio-temporal correlation and feature correlation between the graph node data.

8. The method for early warning of marine ecological disasters based on multi-source heterogeneous big data fusion according to claim 1, wherein, The construction of the marine disaster prediction model according to the cross-domain transfer learning method includes: Extracting the network structure and training parameters of the land disaster prediction model as the basic marine disaster prediction model, and embedding a fully connected layer and a domain adversarial network module for marine ecological disaster prediction in the basic marine disaster prediction model to generate the first marine disaster prediction model; Obtaining historical marine ecological disaster data as target domain data, and obtaining historical land disaster data as source domain data, and correcting the parameters of the first marine disaster prediction model by minimizing the feature distribution difference between the source domain data and the target domain data to obtain the second marine disaster prediction model; Performing data augmentation processing on the historical marine ecological disaster data to obtain second historical marine ecological disaster data; Training the second marine disaster prediction model based on the second historical marine ecological disaster data, and updating the weight parameters of the second marine disaster prediction model by optimizing the loss function to obtain a trained marine disaster prediction model.

9. The method for early warning of marine ecological disasters based on multi-source heterogeneous big data fusion according to claim 8, wherein, The performing data augmentation processing on the historical marine ecological disaster data to obtain second historical marine ecological disaster data includes: Adding Gaussian noise with a mean of zero and an adaptive standard deviation to the time series feature data in the historical marine ecological disaster data, and randomly offsetting the spatial position coordinate data of the historical marine ecological disaster data to generate first augmented data containing spatio-temporal perturbations; Constructing a marine ecological data generation model based on the generative adversarial network, and generating second augmented data conforming to the characteristics of the target sea area based on the marine ecological data generation model, wherein the marine ecological data generation model is designed to learn the distribution law of marine ecological data by inputting the feature vectors of real disaster samples; Performing validity verification on the first augmented data and the second augmented data, and merging the first augmented data and the second augmented data that pass the validity verification to obtain the second historical marine ecological disaster data.

10. The method for early warning of marine ecological disasters based on multi-source heterogeneous big data fusion according to claim 1, wherein, The performing real-time early warning of ecological disasters in the target sea area based on the multi-dimensional prediction results includes: Obtaining real-time hydrological data and meteorological data of the target sea area; Inputting the multi-dimensional prediction results, the real-time hydrological data and the meteorological data into a pre-constructed disaster assessment model, and determining the risk level of ecological disasters according to the output results of the model; Obtaining the ecological sensitive areas of the target sea area, and determining the threatened degree of the ecological sensitive areas based on the disaster impact range in the multi-dimensional prediction results; Executing the early warning control instruction generated by the risk level and the threatened degree.

Images