Marine disaster data processing method based on remote sensing monitoring

Abstract

The present invention relates to the field of marine disaster data processing technology, and in particular to a marine disaster data processing method based on remote sensing monitoring, comprising the following steps: S1: collecting data on ocean temperature, salinity, sea level and marine surface biological activities; S2: performing preliminary processing on the data collected in S1; S3: transmitting the processed data to a ground processing center in real time through a satellite communication system; S4: performing advanced analysis on the data received in S3, identifying and predicting minor changes in the marine environment; S5: predicting the disaster risk within the next 48 hours; S6: automatically generating and publishing early warning information based on the prediction results of S5. The present invention significantly improves the real-time and accuracy of marine disaster early warnings by integrating remote sensing technology, data processing algorithms and information release systems, effectively enhances the timeliness and effectiveness of disaster response measures, and reduces the impact of disasters on human society and the environment.

Description

Technical Field

[0001] The present invention relates to the technical field of marine disaster data processing, and in particular to a marine disaster data processing method based on remote sensing monitoring. Background Art

[0002] In recent years, with climate change and increased human activities, the frequency and destructiveness of marine disasters have increased significantly, such as tsunamis, storm surges and red tides. These disasters not only cause huge damage to the ecological environment of coastal areas, but also seriously threaten the lives and property of local residents. Traditional marine disaster monitoring methods mainly rely on ground stations and ship observations. These methods have great limitations in spatial coverage and real-time performance. In addition, although some existing remote sensing monitoring technologies have improved the efficiency of data collection to a certain extent, they still have obvious deficiencies in the accuracy of data processing and disaster prediction.

[0003] Existing technologies mainly face the following difficulties: first, the data collection method is single and cannot fully capture the multi-dimensional information of the marine environment; second, the data processing and analysis methods lack intelligence and cannot identify and predict small changes in the marine environment in real time; finally, the early warning information release channels are limited, and the information transmission speed and coverage are insufficient, making it impossible to promptly and effectively notify residents and relevant departments in the affected areas.

[0004] Therefore, there is an urgent need for a comprehensive marine disaster data processing method based on remote sensing monitoring, which can realize the real-time collection, efficient processing and accurate prediction of multi-source data, and quickly release early warning information through multiple channels to improve the timeliness and accuracy of disaster response and reduce the losses caused by disasters. Summary of the invention

[0005] Based on the above purpose, the present invention provides a method for processing marine disaster data based on remote sensing monitoring.

[0006] The method for processing marine disaster data based on remote sensing monitoring includes the following steps:

[0007] S1: A multi-modal sensor system on board the satellite, including synthetic aperture radar, infrared thermal imager and high-resolution optical imager, simultaneously collects data on ocean temperature, salinity, sea level height and marine surface biological activity;

[0008] S2: Apply edge computing technology on the satellite to perform preliminary processing on the data collected in S1, including data compression and feature extraction;

[0009] S3: The processed data in S2 is transmitted to the ground processing center in real time via the satellite communication system. The transmission process uses blockchain technology to ensure the security and integrity of the data.

[0010] S4: Performs advanced analysis of data received from S3 using a preset deep learning model at the ground processing center to identify and predict subtle changes in the ocean environment;

[0011] S5: Combine historical disaster data with the analysis results of S4, use the improved Bayesian network model to predict the disaster risk within the next 48 hours, and express the risk level;

[0012] S6: Based on the prediction results of S5, early warning information is automatically generated and issued, including the type of disaster, affected area, expected impact and recommended preventive measures, and the results are visualized through the geographic information system.

[0013] Furthermore, the S1 specifically includes:

[0014] S11: A synthetic aperture radar is installed on the satellite to collect sea surface height data. The synthetic aperture radar transmits microwaves and receives their echoes, calculating the round-trip time of the microwaves, thereby accurately measuring the fluctuations of the sea surface.

[0015] S12: It is also equipped with an infrared thermal imager to measure the ocean surface temperature. The infrared thermal imager captures the infrared radiation emitted by the sea surface and estimates the temperature based on the radiation intensity.

[0016] S13: Equipped with a high-resolution optical imager, it is used to monitor ocean salinity and surface biological activities. By analyzing the spectral characteristics of light reflected from the ocean surface, it can infer changes in salinity and the distribution of algae.

[0017] S14: The above sensor system synchronously processes the collected data through a high-speed data processor, and ensures that the time tag of the data is accurate through the time synchronization system on the satellite.

[0018] Furthermore, the S2 specifically includes:

[0019] S21: An onboard edge computing unit that includes a multi-core processor and a dedicated AI acceleration chip to process the multimodal sensor data collected in S1.

[0020] S22: Perform preliminary processing on the sea surface height data collected by SAR. Specifically, the block compression algorithm is used to compress the original data by sea area, reducing data redundancy. The features retained during the compression process include the amplitude and frequency of sea surface height changes.

[0021] S23: Extract features from the ocean surface temperature data collected by the infrared thermal imager. Specifically, a convolutional neural network model is used to extract features from the temperature image, including temperature gradients and abnormal temperature rise areas.

[0022] S24: Spectral analysis of salinity and biological activity data collected by the high-resolution optical imager was performed, and the frequency band information in the reflectance spectrum was extracted using the fast Fourier transform method to identify the high-frequency characteristics of salinity changes and biological activities, laying the foundation for subsequent analysis;

[0023] S25: Integrate the data processed in S22, S23 and S24, and compress them using a data compression algorithm to generate a data packet to be transmitted.

[0024] Furthermore, the S3 specifically includes:

[0025] S31: configuring a data transmission module in the satellite communication system, wherein the data transmission module includes a high-speed data modem and an encryption chip for processing the data packet generated in S2;

[0026] S32: Uses an efficient data modem to convert data packets into a signal format suitable for satellite communications and sends them to the ground receiving station via a satellite antenna. During the transmission process, the data link layer protocol is used to ensure the order and integrity of the data packets.

[0027] S33: During the transmission process, blockchain technology is used to protect the data packets. Specifically, before the data packets are generated and sent, a unique hash value for each data packet is generated through a hash function, and the hash value is stored in the distributed ledger of the blockchain. This ensures that when each data packet is received, the ground processing center can verify the integrity and non-tampering of the data packet by comparing the hash value;

[0028] S34: At the ground receiving station, a receiving device compatible with the satellite communication module is configured to receive the data packets transmitted from the satellite and demodulate them back to the original data format. The receiving device also includes a blockchain node that can verify the hash value of each data packet in real time;

[0029] S35: Finally, after the ground processing center receives the complete data packet and verifies its integrity and security, it stores the data packet in the local data storage system.

[0030] Furthermore, the S4 specifically includes:

[0031] S41: Configure a high-performance computing cluster in the ground processing center, which includes multiple servers and graphics processing units to run the preset deep learning model;

[0032] S42: Importing the data packet received and verified by S3 into the input interface of the deep learning model, wherein the deep learning model includes multiple convolutional neural network layers and recurrent neural network layers, which are used to process different types of data features; specifically, in the convolutional neural network layer, applying convolution operations to extract spatial features in ocean temperature, salinity, sea surface height and biological activity data; the features include sea surface temperature gradient, salinity distribution pattern, sea surface height change and spatiotemporal distribution of biological activities; in the recurrent neural network layer, processing the features extracted from the convolutional neural network layer, analyzing the time series changes of the data, and the recurrent neural network layer can capture the time dependency in the data, including the periodic changes in temperature and salinity and the short-term fluctuations in the marine environment;

[0033] S43: Combining the outputs of the convolutional neural network layer and the recurrent neural network layer, the deep learning model will generate a multi-dimensional feature vector to represent the current state of the marine environment and identify subtle changes in the marine environment by comparing real-time data with historical data.

[0034] Furthermore, the S43 specifically includes:

[0035] S431: Perform dimensionality reduction operation on the output feature map of the convolutional neural network layer to obtain a spatial feature vector. Specifically, assume that the output of the convolutional neural network layer is a three-dimensional tensor X cnn , whose dimension is: H×W×C, where H represents the height of the feature map, W represents the width of the feature map, and C represents the number of channels of the feature map. It is reduced to a one-dimensional vector V through the global average pooling layer. spatial , the formula is:

[0036] Among them, X cnn (i, j, :) represents all channel values ​​of the feature map at position (i, j);

[0037] S432: Perform a time aggregation operation on the output sequence of the recurrent neural network layer to obtain a time feature vector. First, assume that the output of the recurrent neural network layer is a time series matrix X rnn , whose dimension is: T×D, where T represents the number of time steps and D represents the feature dimension of each time step; then the last hidden state h of the long short-term memory network is t As the time feature vector V temporal , the formula is: V temporal =h t , where h t is the last hidden state at time step t, representing the overall characteristics of the time series data;

[0038] S433: The spatial feature vector V spatial and the time feature vector V temporalSplice to form a comprehensive feature vector V combined , the splicing operation formula is: V combined =[V spatial ; V temporal ], where [;] represents the vector concatenation operation, and the generated V combined The dimension is (C+D);

[0039] S434: Comprehensive feature vector V through the fully connected layer combined Processing is performed to generate the final multi-dimensional feature vector V final , used to represent the current state of the marine environment, the formula is:

[0040] V final =σ(W·V combined +b), where W is the weight matrix with dimension F×(C+D), b is the bias vector with dimension F; σ is the activation function, and F represents the dimension of the final feature vector;

[0041] S435: The final multi-dimensional feature vector V final and the historical data feature vector V historical Compare and calculate the difference between the two vectors. The formula is:

[0042] D change =||V final -V historical ||2, where V historical represents the feature vector of historical data, ||·||2 represents the Euclidean distance, which is used to quantify the difference between the current state and historical data;

[0043] S436: Based on the calculated difference D change , set the threshold θ to determine whether there is a small change, when D change When ≥θ, it is judged that there is a significant change.

[0044] Furthermore, the S5 specifically includes:

[0045] S51: configuring an improved Bayesian network model in the ground processing center, the improved Bayesian network model is pre-trained to include historical disaster data and related environmental characteristics, the model structure includes nodes and edges, the nodes represent different environmental variables, and the edges represent conditional dependencies between the variables;

[0046] S52: The multi-dimensional feature vector V of the current ocean environment generated in S4 final Input into the Bayesian network model, calculate the conditional probability of each node by updating the probability distribution of the node and combining the current features and historical data;

[0047] S53: Based on the output of the Bayesian network model, the disaster risk within the next 48 hours is predicted, and the probability distribution of each disaster type is generated. The disaster types with high probability are marked as potential risks;

[0048] S54: Potential disaster risks are graded and divided into different levels, including low risk, medium risk and high risk, according to the probability value output by the model.

[0049] Furthermore, the S51 specifically includes:

[0050] S511: First, configure a high-performance computing cluster in the ground processing center to build and train the Bayesian network model. The computing cluster includes multiple servers and efficient parallel processing units to ensure the efficiency of model training and inference;

[0051] S512: Build a Bayesian network model and define the nodes and edges of the model. The nodes represent different environmental variables, including ocean temperature, salinity, sea level height, and biological activity; the edges represent the conditional dependencies between variables.

[0052] S513: defining a conditional probability table for each node, specifically a conditional probability table for the node ocean temperature T, which is used to represent the probability distribution of the ocean temperature under different environmental conditions, expressed as P(T|Pa(T)), where Pa(T) represents the parent node set of the node T;

[0053] S514: using a maximum expectation algorithm to estimate the parameters of the Bayesian network model. The maximum expectation algorithm maximizes the likelihood function of the observed data by iteratively optimizing the parameters given the initial parameters and data. The iterative process includes an expectation step and a maximization step. The expectation step is used to calculate the expected value of the unobserved variable given the current parameters. The maximization step is used to update the parameters so that the likelihood function reaches the maximum under the expected value calculated in the expectation step.

[0054] S515: Finally, the model is cross-validated to evaluate the prediction performance of the model and avoid overfitting by dividing the historical data into a training set and a validation set.

[0055] Furthermore, the S53 specifically includes:

[0056] S531: First obtain the multi-dimensional feature vector V of the current ocean environment from the Bayesian network model current , the feature vector includes the current state of ocean temperature T, salinity S, sea surface height H and biological activity B;

[0057] S532: The conditional probability of each disaster type is calculated using the Bayesian network model. The preset disaster types include tsunami D1, storm surge D2, and red tide D3. The Bayesian network model calculates the conditional probability of each disaster type using the following formula: Where P(D i ) is the disaster type D i The prior probability, P(V current |D i ) is in disaster type D i Current environment characteristics V current The conditional probability, P(V current ) is the current environment feature V current The marginal probability of

[0058] S533: Calculate the conditional probability distribution of each disaster type through Bayesian inference, preset the conditional probability table of known nodes and edges of the Bayesian network model, and use the Bayesian formula combined with the current feature vector V current Update the probability distribution of each node;

[0059] S534: For each disaster type, calculate its probability of occurrence within the next 48 hours. Specifically, use the dynamic Bayesian network to expand the time series data into multiple time slices and perform probability inference through the forward algorithm. The specific formula is: in, Indicates the disaster type D at time t+1 i The probability of Denotes the disaster type D at time t+1 i , Denotes the disaster type D at time t j , Denotes the disaster type D at time t j Disaster type D at time t+1 i The conditional probability of Indicates the current environment characteristics V current Next time t disaster type D j The conditional probability of

[0060] S535: Based on the calculation results, a probability distribution diagram for each disaster type is generated to display the probability of occurrence of different disaster types in the next 48 hours, and potential risks are marked by high and low probability values.

[0061] Furthermore, the S6 specifically includes:

[0062] S61: Collect and integrate the prediction results of the Bayesian network model in step S5 to determine the probability and risk level of each disaster type, including tsunamis, storm surges, and red tides;

[0063] S62: Automatically generate warning information based on the prediction results, including the disaster type, affected area, expected impact, and recommended preventive measures;

[0064] S63: Input the generated warning information into the geographic information system and perform visualization of the results. The specific steps include:

[0065] First, the warning information is converted into a GIS-compatible data format, including geographic coordinates, regional polygons, and risk level attributes;

[0066] Then: Map the affected area on a GIS platform, using different colors and markers to represent different hazard types and risk levels;

[0067] Finally: Combine real-time monitoring data and forecast results to dynamically update map information;

[0068] S64: Release warning information through multiple channels to ensure wide dissemination of information, including mobile applications and social media.

[0069] Beneficial effects of the present invention:

[0070] The present invention realizes the real-time collection, efficient processing and accurate prediction of marine disaster data by integrating multimodal sensor system, edge computing, blockchain technology, deep learning model and Bayesian network model. Specifically, the multimodal sensor system can comprehensively capture the multi-dimensional information of the marine environment, edge computing technology performs preliminary data processing on the satellite, significantly improves the data transmission efficiency, blockchain technology ensures the security and integrity of the data transmission process, the deep learning model can accurately identify and analyze the slight changes in the marine environment, and the Bayesian network model combines historical data for risk prediction, which greatly improves the accuracy of disaster warning.

[0071] The present invention uses the geographic information system GIS to visualize the results and releases warning information through multiple channels to ensure the wide dissemination and timely communication of warning information. The warning information covers in detail the type of disaster, affected area, expected impact and recommended preventive measures, providing a reliable reference for decision makers and the public. Combining these technical advantages effectively improves the efficiency and accuracy of marine disaster monitoring and early warning, enhances the ability to respond to disasters, helps to take effective preventive measures in advance, reduces the losses caused by disasters, and protects the lives and property of residents in coastal areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0073] Figure 1 A schematic diagram of a method for processing marine disaster data according to an embodiment of the present invention;

[0074] Figure 2 A schematic diagram of a configuration-improved Bayesian network model process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0076] like Figure 1-Figure 2 As shown, the method for processing marine disaster data based on remote sensing monitoring includes the following steps:

[0077] S1: A multi-modal sensor system on board the satellite, including synthetic aperture radar (SAR), infrared thermal imager and high-resolution optical imager, simultaneously collects data on ocean temperature, salinity, sea surface height and marine surface biological activity;

[0078] S2: Apply edge computing technology on the satellite to perform preliminary processing on the data collected in S1, including data compression and feature extraction, to optimize data transmission efficiency and reduce transmission latency;

[0079] S3: The processed data in S2 is transmitted to the ground processing center in real time via the satellite communication system. The transmission process uses blockchain technology to ensure the security and integrity of the data.

[0080] S4: Performs advanced analysis of data received from S3 using a preset deep learning model at the ground processing center to identify and predict subtle changes in the ocean environment;

[0081] S5: Combine historical disaster data with the analysis results of S4, use the improved Bayesian network model to predict the disaster risk within the next 48 hours, and express the risk level;

[0082] S6: Based on the prediction results of S5, early warning information is automatically generated and issued, including disaster type, affected area, expected impact and recommended preventive measures. At the same time, the results are visualized through the Geographic Information System (GIS) to ensure rapid communication of information and decision support.

[0083] S1 specifically includes:

[0084] S11: A synthetic aperture radar (SAR) is configured on the satellite to collect sea surface height data. The SAR transmits microwaves and receives their echoes, calculating the round-trip time of the microwaves, thereby accurately measuring the fluctuations of the sea surface for monitoring marine disasters such as tsunamis or storm surges.

[0085] S12: It is also equipped with an infrared thermal imager to measure the ocean surface temperature. The infrared thermal imager captures the infrared radiation emitted by the sea surface and estimates the temperature based on the radiation intensity. This data is crucial for analyzing ocean warm currents, ocean current changes and their impact on climate.

[0086] S13: Equipped with a high-resolution optical imager, it is used to monitor ocean salinity and surface biological activity. By analyzing the spectral characteristics of light reflected from the ocean surface, it can infer changes in salinity and the distribution of algae, which is of great significance for assessing ecosystem health and biodiversity.

[0087] S14: The above sensor system synchronously processes the collected data through a high-speed data processor, and ensures that the time tag of the data is accurate through the time synchronization system on the satellite to support precise analysis and real-time monitoring of subsequent data; the above steps not only improve the accuracy and scope of data collection, but also ensure the real-time and reliability of the data, and provide support for rapid response to marine disasters. In this way, the ability of marine disaster warning and research can be greatly enhanced, the losses caused by disasters can be reduced, and high-quality data can be provided for scientific research.

[0088] S2 specifically includes:

[0089] S21: An onboard edge computing unit that includes a multi-core processor and a dedicated AI acceleration chip to process the multimodal sensor data collected in S1.

[0090] S22: Perform preliminary processing on the sea surface height data collected by SAR. Specifically, the block compression algorithm is used to compress the original data by sea area, reduce data redundancy, ensure the efficiency and effectiveness of data transmission, and retain the features in the compression process, including the amplitude and frequency of sea surface height changes;

[0091] S23: Extract features from the ocean surface temperature data collected by the infrared thermal imager. Specifically, a convolutional neural network (CNN) model is used to extract features from the temperature image, including temperature gradients and abnormal temperature rise areas, to provide concise and useful data for subsequent deep learning analysis.

[0092] S24: Spectral analysis of salinity and biological activity data collected by the high-resolution optical imager was performed, and the frequency band information in the reflectance spectrum was extracted using the fast Fourier transform (FFT) method to identify the high-frequency characteristics of salinity changes and biological activities, laying the foundation for subsequent analysis;

[0093] S25: Integrate the data processed in S22, S23 and S24, and compress them through a data compression algorithm to generate data packets to be transmitted; by applying the above-mentioned edge computing technology on satellites, the efficiency of data processing and transmission can be significantly improved, reducing the burden on ground processing centers while ensuring the integrity and practicality of key data. This method enhances real-time monitoring capabilities and improves the overall performance of the marine disaster data processing system.

[0094] S3 specifically includes:

[0095] S31: configuring a data transmission module in the satellite communication system, the data transmission module including a high-speed data modem and an encryption chip for processing the data packet generated by S2;

[0096] S32: Uses an efficient data modem to convert data packets into a signal format suitable for satellite communications and sends them to the ground receiving station via a satellite antenna. During the transmission process, the data link layer protocol is used to ensure the order and integrity of the data packets.

[0097] S33: During the transmission process, blockchain technology is used to protect the data packets. Specifically, before the data packets are generated and sent, a unique hash value for each data packet is generated through a hash function, and the hash value is stored in the distributed ledger of the blockchain. This ensures that when each data packet is received, the ground processing center can verify the integrity and non-tampering of the data packet by comparing the hash value;

[0098] S34: At the ground receiving station, a receiving device compatible with the satellite communication module is configured to receive the data packets transmitted from the satellite and demodulate them back to the original data format. The receiving device also includes a blockchain node that can verify the hash value of each data packet in real time;

[0099] S35: Finally, when the ground processing center receives the complete data packet and verifies its integrity and security, it stores the data packet in the local data storage system, ready for subsequent in-depth analysis and processing; through the above steps, it ensures that the data packet processed in step S2 is safe and correct during the transmission process, and uses blockchain technology to ensure the security and integrity of data transmission, prevent the data packet from being tampered with or lost, and provide a reliable real-time data transmission mechanism.

[0100] S4 specifically includes:

[0101] S41: Configure a high-performance computing cluster in the ground processing center, which includes multiple servers and graphics processing units (GPUs) to run the preset deep learning model, which has been trained on marine environment data;

[0102] S42: Import the data packet received and verified by S3 into the input interface of the deep learning model. The deep learning model includes multiple convolutional neural network (CNN) layers and recurrent neural network (RNN) layers, which are used to process different types of data features. Specifically, in the convolutional neural network layer, convolution operations are applied to extract spatial features in ocean temperature, salinity, sea surface height and biological activity data. The features include sea surface temperature gradient, salinity distribution pattern, sea surface height change and spatiotemporal distribution of biological activities. In the recurrent neural network (RNN) layer, the features extracted from the convolutional neural network layer are processed to analyze the time series changes of the data. The recurrent neural network layer can capture the time dependency in the data, including the periodic changes in temperature and salinity and the short-term fluctuations in the marine environment.

[0103] S43: By integrating the outputs of the convolutional neural network layer and the recurrent neural network layer, the deep learning model will generate a multi-dimensional feature vector to represent the current state of the marine environment, and identify subtle changes in the marine environment by comparing real-time data with historical data. By using a high-performance computing cluster and a preset deep learning model in the ground processing center, it can efficiently perform advanced analysis of marine environmental data, accurately identify and predict subtle environmental changes. Combining the advantages of convolutional neural networks and recurrent neural networks, the model can capture the spatial and temporal characteristics of marine data, significantly improving the accuracy and real-time nature of disaster warnings and enhancing the ability to respond to marine disasters.

[0104] S43 specifically includes:

[0105] S431: Perform dimensionality reduction operation on the output feature map of the convolutional neural network layer to obtain a spatial feature vector. Specifically, assume that the output of the convolutional neural network layer is a three-dimensional tensor X cnn , whose dimension is: H×W×C, where H represents the height of the feature map, W represents the width of the feature map, and C represents the number of channels of the feature map. It is reduced to a one-dimensional vector V through the Global Average Pooling (GAP) layer. spatial , the formula is: Among them, X cnn (i, j, :) represents all channel values ​​of the feature map at position (i, j);

[0106] S432: Perform a time aggregation operation on the output sequence of the recurrent neural network (RNN) layer to obtain a time feature vector. First, assume that the output of the recurrent neural network layer is a time series matrix X rnn, whose dimension is: T×D, where T represents the number of time steps and D represents the feature dimension of each time step; then the last hidden state h of the long short-term memory network (LSTM) is t As the time feature vector V temporal , the formula is: V temporal =h t , where h t is the last hidden state at time step t, representing the overall characteristics of the time series data;

[0107] S433: The spatial feature vector V spatial and the time feature vector V temporal Splice to form a comprehensive feature vector V combined， The splicing operation formula is: V combined =[V temporal ; V temporal ], where [;] represents the vector concatenation operation, and the generated V combined The dimension is (C+D);

[0108] S434: Through the fully connected layer (FC), the comprehensive feature vector V combined Processing is performed to generate the final multi-dimensional feature vector V final , used to represent the current state of the marine environment, the formula is: V final =σ(W·V combined +b), where W is the weight matrix with dimension F×(C+D), b is the bias vector with dimension F; σ is the activation function, and F represents the dimension of the final feature vector;

[0109] S435: The final multi-dimensional feature vector V final and the historical data feature vector V historical Compare and calculate the difference between the two vectors. The formula is:

[0110] D change =||V final -V historical ||2, where V historical represents the feature vector of historical data, ||·||2 represents the Euclidean distance, which is used to quantify the difference between the current state and historical data;

[0111] S436: Based on the calculated difference D change , set the threshold θ to determine whether there is a small change, when D changeWhen ≥θ, it is determined that there is a significant change. Through the above detailed feature vector generation and comparison steps, the spatial and temporal features extracted by the convolutional neural network and the recurrent neural network are used to generate a multi-dimensional feature vector that accurately represents the current state of the marine environment. By comparing real-time data with historical data, small changes in the marine environment can be effectively identified, thereby improving the accuracy and response speed of marine disaster warnings and ensuring that corresponding preventive measures are taken in a timely manner.

[0112] S5 specifically includes:

[0113] S51: configuring an improved Bayesian network model in the ground processing center, the improved Bayesian network model is pre-trained to include historical disaster data and related environmental characteristics, the model structure includes nodes and edges, the nodes represent different environmental variables, and the edges represent conditional dependencies between the variables;

[0114] S52: The multi-dimensional feature vector V of the current ocean environment generated in S4 final Input into the Bayesian network model, calculate the conditional probability of each node by updating the probability distribution of the node and combining the current features and historical data;

[0115] S53: Based on the output of the Bayesian network model, the disaster risk within the next 48 hours is predicted, and the probability distribution of each disaster type is generated. The disaster types with high probability are marked as potential risks;

[0116] S54: Potential disaster risks are graded and divided into different levels, including low risk, medium risk and high risk, according to the probability value output by the model.

[0117] S51 specifically includes:

[0118] S511: First, configure a high-performance computing cluster in the ground processing center to build and train the Bayesian network model. The computing cluster includes multiple servers and efficient parallel processing units (such as GPUs) to ensure the efficiency of model training and inference;

[0119] S512: Build a Bayesian network model and define the nodes and edges of the model. The nodes represent different environmental variables, including ocean temperature, salinity, sea level and biological activity; the edges represent the conditional dependencies between variables. The initial structure of the model is designed based on expert knowledge and historical data.

[0120] S513: defining a conditional probability table (CPT) for each node, specifically a conditional probability table for the node ocean temperature T, which is used to represent the probability distribution of the ocean temperature under different environmental conditions, expressed as P(T|Pa(T)), where Pa(T) represents the parent node set of the node T, and the conditional probability table is estimated by statistically analyzing historical data;

[0121] S514: Parameter estimation of the Bayesian network model is performed using the maximum expectation (EM) algorithm. The maximum expectation algorithm maximizes the likelihood function of the observed data by iteratively optimizing the parameters given the initial parameters and data. The iterative process includes an expectation step (E-step) and a maximization step (M-step); the expectation step is used to calculate the expected value of the unobserved variable given the current parameters; the maximization step is used to update the parameters so that the likelihood function reaches the maximum under the expected value calculated in the expectation step;

[0122] S515: Finally, the model is cross-validated. The prediction performance of the model is evaluated by dividing the historical data into training set and validation set. Cross-validation helps to adjust the model structure and parameters to avoid overfitting. By configuring and training the improved Bayesian network model in the ground processing center, the conditional dependency relationship between environmental variables can be accurately constructed. The prediction accuracy and reliability of the model can be improved through the statistics and optimization of historical data. The high-performance computing cluster and the maximum expectation algorithm are used to ensure the efficiency and effectiveness of model training, laying a solid foundation for the accurate prediction of subsequent disaster risks.

[0123] S53 specifically includes:

[0124] S531: First obtain the multi-dimensional feature vector V of the current ocean environment from the Bayesian network model current , the feature vector includes the current state of ocean temperature T, salinity S, sea surface height H and biological activity B;

[0125] S532: The conditional probability of each disaster type is calculated using the Bayesian network model. The preset disaster types include tsunami D1, storm surge D2, and red tide D3. The Bayesian network model calculates the conditional probability of each disaster type using the following formula: Where P(D i ) is the disaster type D i The prior probability, P(V current |D i ) is in disaster type D i Current environment characteristics V current The conditional probability, P(V current ) is the current environment feature V current The marginal probability of

[0126] S533: Calculate the conditional probability distribution of each disaster type through Bayesian inference, preset the conditional probability table of known nodes and edges of the Bayesian network model, and use the Bayesian formula combined with the current feature vector V current Update the probability distribution of each node;

[0127] S534: For each disaster type, calculate its probability of occurrence within the next 48 hours. Specifically, use the dynamic Bayesian network (DBN) to expand the time series data into multiple time slices, and use the forward algorithm to perform probability inference. The specific formula is: in, Indicates the disaster type D at time t+1 i The probability of Denotes the disaster type D at time t+1 i , Denotes the disaster type D at time t j , Denotes the disaster type D at time t j Disaster type D at time t+1 i The conditional probability of Indicates the current environment characteristics V current Next time t disaster type D j The conditional probability of

[0128] S535: Based on the calculation results, a probability distribution map of each disaster type is generated to show the probability of occurrence of different disaster types in the next 48 hours, and potential risks are marked by high and low probability values ​​to support decision-making and warning issuance; through the above steps, the Bayesian network model is used for probability calculation and inference, which can accurately predict the disaster risk in the next 48 hours. The dynamic Bayesian network is used to expand and infer the time series data, which improves the time accuracy and reliability of the prediction. The generated probability distribution map intuitively shows the potential risks of various disasters, provides an important reference for decision makers, and helps to take preventive measures in advance to reduce the losses caused by disasters.

[0129] S6 specifically includes:

[0130] S61: Collect and integrate the prediction results of the Bayesian network model in step S5 to determine the probability and risk level of each disaster type, including tsunamis, storm surges, and red tides;

[0131] S62: Automatically generate warning information based on the prediction results, including disaster type, affected area, estimated impact, and recommended preventive measures; among them, the disaster type is used to identify the specific type of disaster, including tsunami, storm surge, or red tide; the affected area is based on the geographical scope predicted by the model, which is used to identify the specific area that is potentially affected; the estimated impact is to assess the impact based on the disaster type and probability of occurrence, and the impact is divided into three levels of risk: low, medium, and high; recommended preventive measures are used to provide specific preventive suggestions for each type of disaster, including evacuation routes, shelter locations, and protective measures;

[0132] S63: Input the generated warning information into the geographic information system and perform visualization of the results. The specific steps include:

[0133] First, the warning information is converted into a GIS-compatible data format, including geographic coordinates, regional polygons, and risk level attributes;

[0134] Then: Map the affected area on a GIS platform, using different colors and markers to represent different hazard types and risk levels;

[0135] Finally: Combine real-time monitoring data and forecast results to dynamically update map information to ensure the timeliness and accuracy of early warning information;

[0136] S64: Release warning information through multiple channels to ensure that the information is widely disseminated, including mobile applications and social media. Through the above steps, detailed warning information can be automatically generated and released, including disaster type, affected area, expected impact and recommended prevention measures. At the same time, the results are visualized using the geographic information system (GIS) to make the warning information more intuitive and easy to understand. Multi-channel information release ensures the wide dissemination and timely communication of warning information, provides a reliable reference for the public and decision makers, and helps to quickly take effective prevention measures to reduce the losses caused by disasters.

Claims

1. A method for processing marine disaster data based on remote sensing monitoring, characterized in that: The following steps are involved: S1: A multi-modal sensor system on board the satellite, including synthetic aperture radar, infrared thermal imager and high-resolution optical imager, simultaneously collects data on ocean temperature, salinity, sea level height and marine surface biological activity; S2: Apply edge computing technology on the satellite to perform preliminary processing on the data collected in S1, including data compression and feature extraction; S3: The processed data in S2 is transmitted to the ground processing center in real time via the satellite communication system. The transmission process uses blockchain technology to ensure the security and integrity of the data. S4: Performs advanced analysis of data received from S3 using a preset deep learning model at the ground processing center to identify and predict subtle changes in the ocean environment; The S4 specifically includes: S41: Configure a high-performance computing cluster in the ground processing center, which includes multiple servers and graphics processing units to run the preset deep learning model; S42: Importing the data packet received and verified by S3 into the input interface of the deep learning model, wherein the deep learning model includes multiple convolutional neural network layers and recurrent neural network layers, which are used to process different types of data features; specifically, in the convolutional neural network layer, applying convolution operations to extract spatial features in ocean temperature, salinity, sea surface height and biological activity data; the features include sea surface temperature gradient, salinity distribution pattern, sea surface height change and spatiotemporal distribution of biological activities; in the recurrent neural network layer, processing the features extracted from the convolutional neural network layer, analyzing the time series changes of the data, and the recurrent neural network layer can capture the time dependency in the data, including the periodic changes in temperature and salinity and the short-term fluctuations in the marine environment; S43: Combining the outputs of the convolutional neural network layer and the recurrent neural network layer, the deep learning model will generate a multi-dimensional feature vector to represent the current state of the marine environment and identify subtle changes in the marine environment by comparing real-time data with historical data; The S43 specifically includes: S431: Perform dimensionality reduction operation on the output feature map of the convolutional neural network layer to obtain a spatial feature vector. Specifically, assume that the output of the convolutional neural network layer is a three-dimensional tensor X cnn , whose dimension is: H×W×C, where H represents the height of the feature map, W represents the width of the feature map, and C represents the number of channels of the feature map. It is reduced to a one-dimensional vector V through the global average pooling layer. spatial , the formula is: Among them, X cnn (i, j, :) represents all channel values ​​of the feature map at position (i, j); S432: Perform a time aggregation operation on the output sequence of the recurrent neural network layer to obtain a time feature vector. First, assume that the output of the recurrent neural network layer is a time series matrix X rnn , whose dimension is: T×D, where T represents the number of time steps and D represents the feature dimension of each time step; then the last hidden state h of the long short-term memory network is t As the time feature vector V temporal , the formula is: V temporal =h t , where h t is the last hidden state at time step t, representing the overall characteristics of the time series data; S433: The spatial feature vector V spatial and the time feature vector V temporal Splice to form a comprehensive feature vector V combined , the splicing operation formula is: V combined =[V spatial ; V temporal ], where [;] represents the vector concatenation operation, and the generated V combined The dimension is (C+D); S434: Comprehensive feature vector V through the fully connected layer combined Processing is performed to generate the final multi-dimensional feature vector V final , used to represent the current state of the marine environment, the formula is: V final =σ(W·V combined +b), where W is the weight matrix with dimension F×(C+D), b is the bias vector with dimension F; σ is the activation function, and F represents the dimension of the final feature vector; S435: The final multi-dimensional feature vector V final and the historical data feature vector V historical Compare and calculate the difference between the two vectors. The formula is: D change =||V final -V historical ||2, where V historical represents the feature vector of historical data, ∥·∥2 represents the Euclidean distance, which is used to quantify the difference between the current state and historical data; S436: Based on the calculated difference D change , set the threshold θ to determine whether there is a small change, when D change When ≥θ, it is judged that there is a significant change; S5: Combine historical disaster data with the analysis results of S4, use the improved Bayesian network model to predict the disaster risk within the next 48 hours, and express the risk level; S6: Based on the prediction results of S5, early warning information is automatically generated and issued, including the type of disaster, affected area, expected impact and recommended preventive measures, and the results are visualized through the geographic information system.

2. The method for processing marine disaster data based on remote sensing monitoring according to claim 1, characterized in that: The S1 specifically includes: S11: A synthetic aperture radar is installed on the satellite to collect sea surface height data. The synthetic aperture radar transmits microwaves and receives their echoes, calculating the round-trip time of the microwaves, thereby accurately measuring the fluctuations of the sea surface. S12: It is also equipped with an infrared thermal imager to measure the ocean surface temperature. The infrared thermal imager captures the infrared radiation emitted by the sea surface and estimates the temperature based on the radiation intensity. S13: Equipped with a high-resolution optical imager, it is used to monitor ocean salinity and surface biological activities. By analyzing the spectral characteristics of light reflected from the ocean surface, it can infer changes in salinity and the distribution of algae. S14: The above sensor system synchronously processes the collected data through a high-speed data processor, and ensures that the time tag of the data is accurate through the time synchronization system on the satellite.

3. The method for processing marine disaster data based on remote sensing monitoring according to claim 1, characterized in that: The S2 specifically includes: S21: An onboard edge computing unit that includes a multi-core processor and a dedicated AI acceleration chip to process the multimodal sensor data collected in S1. S22: Perform preliminary processing on the sea surface height data collected by SAR. Specifically, the block compression algorithm is used to compress the original data by sea area, reducing data redundancy. The features retained during the compression process include the amplitude and frequency of sea surface height changes. S23: Extract features from the ocean surface temperature data collected by the infrared thermal imager. Specifically, a convolutional neural network model is used to extract features from the temperature image, including temperature gradients and abnormal temperature rise areas. S24: Spectral analysis of salinity and biological activity data collected by the high-resolution optical imager was performed, and the frequency band information in the reflectance spectrum was extracted using the fast Fourier transform method to identify the high-frequency characteristics of salinity changes and biological activities, laying the foundation for subsequent analysis; S25: Integrate the data processed in S22, S23 and S24, and compress them using a data compression algorithm to generate a data packet to be transmitted.

4. The method for processing marine disaster data based on remote sensing monitoring according to claim 1, characterized in that: The S3 specifically includes: S31: configuring a data transmission module in the satellite communication system, wherein the data transmission module includes a high-speed data modem and an encryption chip for processing the data packet generated in S2; S32: Uses an efficient data modem to convert data packets into a signal format suitable for satellite communications and sends them to the ground receiving station via a satellite antenna. During the transmission process, the data link layer protocol is used to ensure the order and integrity of the data packets. S33: During the transmission process, blockchain technology is used to protect the data packets. Specifically, before the data packets are generated and sent, a unique hash value for each data packet is generated through a hash function, and the hash value is stored in the distributed ledger of the blockchain. This ensures that when each data packet is received, the ground processing center can verify the integrity and non-tampering of the data packet by comparing the hash value; S34: At the ground receiving station, a receiving device compatible with the satellite communication module is configured to receive the data packets transmitted from the satellite and demodulate them back to the original data format. The receiving device also includes a blockchain node that can verify the hash value of each data packet in real time; S35: Finally, after the ground processing center receives the complete data packet and verifies its integrity and security, it stores the data packet in the local data storage system.

5. The method for processing marine disaster data based on remote sensing monitoring according to claim 1, characterized in that: The S5 specifically includes: S51: configuring an improved Bayesian network model in the ground processing center, the improved Bayesian network model is pre-trained to include historical disaster data and related environmental characteristics, the model structure includes nodes and edges, the nodes represent different environmental variables, and the edges represent conditional dependencies between the variables; S52: The multi-dimensional feature vector V of the current ocean environment generated in S4 final Input into the Bayesian network model, calculate the conditional probability of each node by updating the probability distribution of the node and combining the current features and historical data; S53: Based on the output of the Bayesian network model, the disaster risk within the next 48 hours is predicted, and the probability distribution of each disaster type is generated. The disaster types with high probability are marked as potential risks; S54: Potential disaster risks are graded and divided into different levels, including low risk, medium risk and high risk, according to the probability value output by the model.

6. The method for processing marine disaster data based on remote sensing monitoring according to claim 5 is characterized in that: The S51 specifically includes: S511: First, configure a high-performance computing cluster in the ground processing center to build and train the Bayesian network model. The computing cluster includes multiple servers and efficient parallel processing units to ensure the efficiency of model training and inference; S512: Build a Bayesian network model and define the nodes and edges of the model. The nodes represent different environmental variables, including ocean temperature, salinity, sea level height, and biological activity; the edges represent the conditional dependencies between variables. S513: defining a conditional probability table for each node, specifically a conditional probability table for the node ocean temperature T, which is used to represent the probability distribution of the ocean temperature under different environmental conditions, expressed as P(T|Pa(T)), where Pa(T) represents the parent node set of the node T; S514: using a maximum expectation algorithm to estimate the parameters of the Bayesian network model. The maximum expectation algorithm maximizes the likelihood function of the observed data by iteratively optimizing the parameters given the initial parameters and data. The iterative process includes an expectation step and a maximization step. The expectation step is used to calculate the expected value of the unobserved variable given the current parameters. The maximization step is used to update the parameters so that the likelihood function reaches the maximum under the expected value calculated in the expectation step. S515: Finally, the model is cross-validated to evaluate the prediction performance of the model and avoid overfitting by dividing the historical data into a training set and a validation set.

7. The method for processing marine disaster data based on remote sensing monitoring according to claim 6 is characterized in that: The S53 specifically includes: S531: First obtain the multi-dimensional feature vector V of the current ocean environment from the Bayesian network model current , the feature vector includes the current state of ocean temperature T, salinity S, sea surface height H and biological activity B; S532: The conditional probability of each disaster type is calculated using the Bayesian network model. The preset disaster types include tsunami D1, storm surge D2, and red tide D3. The Bayesian network model calculates the conditional probability of each disaster type using the following formula: Where P(D i ) is the disaster type D i The prior probability, P(V current |D i ) is in disaster type D i Current environment characteristics V current The conditional probability, P(V current ) is the current environment feature V current The marginal probability of S533: Calculate the conditional probability distribution of each disaster type through Bayesian inference, preset the conditional probability table of known nodes and edges of the Bayesian network model, and use the Bayesian formula combined with the current feature vector V current Update the probability distribution of each node; S534: For each disaster type, calculate its probability of occurrence within the next 48 hours. Specifically, use the dynamic Bayesian network to expand the time series data into multiple time slices and perform probability inference through the forward algorithm. The specific formula is: in, Indicates the disaster type D at time t+1 i The probability of Indicates the disaster type at time t+1 Indicates the disaster type at time t Denotes the disaster type D at time t j Disaster type D at time t+1 i The conditional probability of Indicates the current environment characteristics V current Next time t disaster type D j The conditional probability of S535: Based on the calculation results, a probability distribution diagram for each disaster type is generated to display the probability of occurrence of different disaster types in the next 48 hours, and potential risks are marked by high and low probability values.

8. The method for processing marine disaster data based on remote sensing monitoring according to claim 7 is characterized in that: The S6 specifically includes: S61: Collect and integrate the prediction results of the Bayesian network model in step S5 to determine the probability and risk level of each disaster type, including tsunamis, storm surges, and red tides; S62: Automatically generate warning information based on the prediction results, including the disaster type, affected area, expected impact, and recommended preventive measures; S63: Input the generated warning information into the geographic information system and perform visualization of the results. The specific steps include: First, the warning information is converted into a GIS-compatible data format, including geographic coordinates, regional polygons, and risk level attributes; Then: Map the affected area on a GIS platform, using different colors and markers to represent different hazard types and risk levels; Finally: Combine real-time monitoring data and forecast results to dynamically update map information; S64: Release warning information through multiple channels to ensure wide dissemination of information, including mobile applications and social media.

Images