Flood monitoring and early warning method and system based on multi-source remote sensing satellite data fusion

Abstract

The invention provides a flood monitoring and early warning method and system based on multi-source remote sensing satellite data fusion. The method comprises the following steps: collecting various types of remote sensing satellite data in a flood area; preprocessing the collected remote sensing satellite data; integrating different scale features of the preprocessed remote sensing satellite data to obtain feature fusion; popularizing feature fusion based on a revisit channel attention module from a same-layer scene to a cross-layer scene; flood recognition is carried out by using the fused remote sensing image, and the boundary and range of a flood area are extracted; according to the identified flood area and the historical flood data, flood early warning is carried out in combination with meteorological and hydrological information, and an early warning result is obtained; and comprehensively analyzing the flood cause, the flood process and the flood influence of the flood event. By integrating multi-source satellite data, the invention aims to improve the accuracy of flood large-range detection and strengthen the prediction capability of flood trend, thereby ensuring the accuracy and reliability of flood space-time early warning.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing geographic information technology, and in particular relates to a flood monitoring and early warning method and system for multi-source remote sensing satellite data fusion. Background Art

[0002] With the acceleration of global climate change and urbanization, flood disasters have occurred frequently, posing a serious threat to people's lives and property. Therefore, timely and accurate acquisition of flood disaster information, early warning and analysis of floods are of great significance to reducing flood disaster losses. Traditional flood warning methods mainly rely on ground observation stations and meteorological data. However, these methods have problems such as limited observation range and untimely data updates. In recent years, the development of remote sensing technology has provided new solutions for flood warning. Multi-source remote sensing satellites can provide large-scale, fast and accurate surface information, providing important data support for flood warning and analysis.

[0003] In recent years, with the rapid development of remote sensing technology, multi-source remote sensing satellites have been used more and more widely in flood warning and analysis. Many scholars and research institutions at home and abroad are committed to the research of flood warning and analysis using remote sensing data. At present, some flood analysis methods and early warning systems based on remote sensing technology have been developed and have achieved certain results in practical applications. These systems mainly use remote sensing data to extract flood characteristic information, and combine ground observation data and meteorological data to warn and analyze floods. However, how to effectively fuse multi-source remote sensing satellite data to improve the accuracy and efficiency of flood warning and analysis is still a problem that needs to be solved urgently. Multi-source remote sensing satellite data has different spatial resolutions, temporal resolutions, and spectral resolutions. How to effectively fuse these data and extract flood characteristic information is one of the key issues that need to be solved at present. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a flood monitoring and early warning method and system based on multi-source remote sensing satellite data fusion, and constructs a revisit channel attention module, aiming to improve the accuracy and timeliness of flood early warning by integrating different types of remote sensing satellite data;

[0005] The technical solution of the method of the present invention is a flood monitoring and early warning method based on multi-source remote sensing satellite data fusion, and the specific steps are as follows:

[0006] Step 1: Collect various types of remote sensing satellite data of the flooded area;

[0007] Step 2: Perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data;

[0008] Step 3: The preprocessed remote sensing satellite data are integrated with features of different scales to obtain better performance feature fusion; the feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario;

[0009] Step 4: Use the fused remote sensing image to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area;

[0010] Step 5: Based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, flood warnings are carried out to obtain warning results;

[0011] Step 6: Conduct a comprehensive analysis of flood causes, flood processes, and flood impacts.

[0012] Preferably, the remote sensing satellite data collection in step 1 needs to ensure the timeliness and accuracy of the data so as to obtain the latest surface information in a timely manner; at the same time, factors such as the coverage and resolution of different remote sensing satellite data need to be considered to ensure the comprehensiveness and reliability of the data.

[0013] Preferably, in step 2, data preprocessing can eliminate radiation errors in remote sensing images through radiation calibration and improve the radiation accuracy of images; geometric correction can eliminate geometric distortion in images and make images more consistent with the real surface morphology; atmospheric correction can eliminate the influence of atmosphere on remote sensing images and improve the clarity and contrast of images; distortion correction formula:

[0014]

[0015] r measured is the satellite's internal parameter, measuring the radial distance; k 1 and k 2 is the radial distortion coefficient.

[0016] Preferably, the data fusion in step 3 is to further improve the quality and resolution of the remote sensing image; a revisit channel attention module is constructed for data fusion;

[0017] The feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario, including: short skip connections and long skip connections, and even internal feature integration;

[0018] The revisit channel attention module includes a multi-scale channel fusion submodule, a local channel context aggregation submodule, and a global context fusion submodule; the input is the data feature map from different satellites, the channel and feature map of size H×W; these inputs are used for subsequent feature extraction and attention weight calculation;

[0019] First, the multi-scale channel fusion submodule performs multi-scale feature fusion on the input multi-source satellite data. Given a source satellite data X, a channel and feature map of size H×W, the channel attention weight w is calculated as follows:

[0020] w=σ(g(X))=σ(B(W 2 δ(B(W 1 (g(X))))));

[0021] where g(X) represents the global feature context, and is the global average pooling; δ represents the ReLU activation function, B represents batch normalization; σ is the Sigmoid function; this is achieved by a link with two fully connected layers, where W 1 is the dimension reduction layer, W 2 It is a dimension-increasing layer; it aggregates local and multi-source data context information;

[0022] The local channel context aggregation submodule realizes the aggregation of local context through the convolution operation PWConv; the key idea is that by changing the spatial pooling size, channel attention can be achieved at multiple scales; the local context needs to be added to the global context in the attention module; point-by-point convolution is selected as the local channel context aggregator, and the point-by-point channel interaction at each spatial position is utilized; the local channel context L(X) is calculated as follows:

[0023] L(X)=B(PWConυ 1 (δ(B(PWCon 1 (X)))))

[0024] Among them, PWConv is a convolution with a convolution kernel of 1×1; L(X) retains and highlights the details in low-level features; finally, the global context fusion submodule combines the local channel context and the global channel context to obtain the global feature representation X of multi-source data g :

[0025]

[0026] Where M(X) represents the generated attention weight; represents broadcast addition, represents element-wise multiplication; N represents multiple satellite data sources.

[0027] Preferably, the flood identification in step 4 is a key step in flood warning and analysis; by adopting image processing techniques such as threshold segmentation, edge detection, and morphological processing, the boundary and range of the flood area can be effectively extracted; at the same time, other auxiliary information such as terrain, landform, meteorological and other data can be combined to more accurately and comprehensively identify and analyze the flood area; the flood identification index NDWI is calculated as follows:

[0028]

[0029] NIR represents the reflectivity of the near infrared band, and SWIR represents the reflectivity of the short-wave infrared band. Deep learning is used to extract flood areas from optical remote sensing images. The input image is defined as I, and its corresponding feature map is defined as F. The goal is to generate a binary mask M, where M ij Indicates whether the pixel (i, j) belongs to the flood area; convolutional layers and pooling layers are used to extract high-level features of the image; the output feature map of the encoder is F enc ;

[0030] Decoder F enc Map it back to the original image size; use deconvolution layers and upsampling operations to restore the resolution; the output of the decoder is F dec ;

[0031] A convolutional layer is used to generate the water body segmentation probability map P, where P ij represents the probability that pixel (i, j) belongs to a water body; the probability map P is compared with a threshold τ to obtain the final binary mask M:

[0032]

[0033] The formula of the entire spinous process deep learning model for extracting flood area from optical remote sensing images can be expressed as:

[0034] M = Threshold(P, τ)

[0035] Threshold is a threshold processing function.

[0036] Preferably, in step 5, flood warning needs to comprehensively consider multiple factors, including historical flood data, meteorological, hydrological and other information; by analyzing and modeling these factors, the probability of flood occurrence, peak flow, arrival time and other information can be predicted, providing an important reference for disaster prevention and response; Flood warning model:

[0037] Flood prediction = Model flood (H, M, W)

[0038] Model floodis a flood prediction model based on Transformer architecture; H is a feature vector of historical flood data, including the frequency of past flood events, peak flow, flood duration, etc.; historical data can be obtained through statistical analysis; M is a feature value of meteorological factors, including rainfall, rainfall intensity, rainfall distribution, etc.; meteorological data can be obtained from meteorological stations, satellites or other meteorological sensors; W is a hydrological factor, including the topography, soil type, river network, water storage capacity, etc. of the basin; hydrological data can be obtained through hydrological models and geographic information systems;

[0039] The warning results in step 5 include: probability of flood occurrence, flood peak flow, and flood arrival time information.

[0040] Preferably, the comprehensive analysis in step 6 is as follows:

[0041] Using geographic information system technology, flood data is overlaid and analyzed with geographic information such as topography, landforms, and population distribution to provide decision support for post-disaster rescue and reconstruction;

[0042] In step 6, flood warning requires analysis of the causes, processes, and impacts of floods, which can provide important experience and lessons for future disaster prevention and response. At the same time, it is also possible to use technical means such as geographic information systems (GIS) to overlay and analyze flood data with related geographic information, providing more scientific and effective decision-making support for post-disaster rescue and reconstruction.

[0043] The technical solution of the system of the present invention is a flood monitoring and early warning system integrating multi-source remote sensing satellite data, comprising:

[0044] A data acquisition module for collecting various types of remote sensing satellite data on flooded areas;

[0045] The preprocessing module is used to perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data;

[0046] The feature fusion module is used to integrate the pre-processed remote sensing satellite data with different scale features to obtain better performance; the feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario;

[0047] The flood area recognition module is used to use the fused remote sensing images to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area;

[0048] The early warning module is used to conduct flood warnings based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, to obtain early warning results;

[0049] The comprehensive analysis module is used for comprehensive analysis of flood causes, flood processes, and flood impacts.

[0050] The present invention proposes a flood warning and analysis method based on the revisit channel attention module to fuse multi-source remote sensing satellite data. Compared with the traditional single remote sensing data source method, it has the following significant advantages and beneficial effects:

[0051] By integrating different types of remote sensing satellite data, the accuracy of large-scale flood detection can be improved, and the ability to predict flood trends can be strengthened, thereby ensuring the accuracy and reliability of flood spatiotemporal warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 : A flow chart of a method according to an embodiment of the present invention.

[0053] Figure 2 : Flood prediction and warning map of an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] In specific implementation, the method proposed in the technical solution of the present invention can be implemented by technical personnel in this field using computer software technology to realize automatic operation process. System devices for implementing the method, such as computer-readable storage media storing the corresponding computer program of the technical solution of the present invention and computer equipment running the corresponding computer program, should also be within the scope of protection of the present invention.

[0056] The following is combined with Figure 1-2 The specific implementation of the present invention is a flood monitoring and early warning method and system based on multi-source remote sensing satellite data fusion, which is as follows:

[0057] The specific implementation of the method of the present invention is a flood monitoring and early warning method based on multi-source remote sensing satellite data fusion, and the specific steps are as follows:

[0058] like Figure 1 FIG. 1 is a flow chart of a method according to an embodiment of the present invention.

[0059] Remote sensing data of flood-affected areas are collected through satellite remote sensing platforms.

[0060] In the process of collecting remote sensing data in flood-affected areas, each step is crucial. The first is the selection of satellites, which needs to be determined according to the specific needs of flood monitoring. For example, for large-scale flood monitoring, satellites with wide coverage areas and high resolution, such as MODIS or Sentinel series satellites, should be selected. Secondly, it is also key to determine the type of remote sensing data required, which includes optical images and synthetic aperture radar (SAR) images. Optical images perform well in clear weather conditions, while SAR images can penetrate clouds and are suitable for cloudy or rainy days.

[0061] Data acquisition is to obtain the required remote sensing data through satellite remote sensing platforms, such as the Earth Observation System Data and Information System (EOSDIS) or the European Space Agency (ESA) Copernicus open access hub. After acquisition, the raw remote sensing data needs to be preprocessed, including radiation correction, atmospheric correction, geometric correction, etc., to improve the accuracy and availability of the data.

[0062] On the basis of data processing, image processing and analysis techniques, such as water index calculation and classification algorithm, are applied to extract flood characteristic information from the processed remote sensing data. Finally, the extracted flood characteristic information is integrated with other relevant data (such as topography, land use, and meteorological data) for comprehensive analysis to more accurately assess the flood situation. This series of steps constitutes the key process of remote sensing data collection in flood-affected areas.

[0063] Step 1: Collect various types of remote sensing satellite data of the flooded area;

[0064] The remote sensing satellite data collection in step 1 needs to ensure the timeliness and accuracy of the data so as to obtain the latest surface information in a timely manner; at the same time, factors such as the coverage and resolution of different remote sensing satellite data need to be considered to ensure the comprehensiveness and reliability of the data.

[0065] Step 2: Perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data;

[0066] In step 2, data preprocessing can eliminate radiation errors in remote sensing images through radiation calibration and improve the radiation accuracy of images; geometric correction can eliminate geometric distortion in images and make images more consistent with the real surface morphology; atmospheric correction can eliminate the influence of atmosphere on remote sensing images and improve the clarity and contrast of images; distortion correction formula:

[0067]

[0068] r measured =0.4 is the satellite's internal parameter, measuring the radial distance; k1 = 0.6 and k 2 =0.3 is the radial distortion coefficient.

[0069] Step 3: The preprocessed remote sensing satellite data are integrated with features of different scales to obtain better performance feature fusion; the feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario;

[0070] The data fusion in step 3 is to further improve the quality and resolution of remote sensing images; a revisit channel attention module is constructed for data fusion;

[0071] The feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario, including: short skip connections and long skip connections, and even internal feature integration;

[0072] The revisit channel attention module includes a multi-scale channel fusion submodule, a local channel context aggregation submodule, and a global context fusion submodule; the input is the data feature map from different satellites, the channel and feature map of size H×W; these inputs are used for subsequent feature extraction and attention weight calculation;

[0073] First, the multi-scale channel fusion submodule performs multi-scale feature fusion on the input multi-source satellite data. Given a source satellite data X, a channel and feature map of size H×W, the channel attention weight ω is calculated as follows:

[0074] w=σ(g(X))=σ(B(W 2 δ(B(W 1 (g(X))))));

[0075] where g(X) represents the global feature context, and is the global average pooling; δ represents the ReLU activation function, B represents batch normalization; σ is the Sigmoid function; this is achieved by a link with two fully connected layers, where W 1 is the dimension reduction layer, W 2 It is a dimension-increasing layer; it aggregates local and multi-source data context information;

[0076] The local channel context aggregation submodule realizes the aggregation of local context through the convolution operation PWConv; the key idea is that by changing the spatial pooling size, channel attention can be achieved at multiple scales; the local context needs to be added to the global context in the attention module; point-by-point convolution is selected as the local channel context aggregator, and the point-by-point channel interaction at each spatial position is utilized; the local channel context L(X) is calculated as follows:

[0077] L(X)=B(PWConυ 1(δ(B(PWCon 1 (X)))))

[0078] Among them, PWConv is a convolution with a convolution kernel of 1×1; L(X) retains and highlights the details in low-level features; finally, the global context fusion submodule combines the local channel context and the global channel context to obtain the global feature representation X of multi-source data g :

[0079]

[0080] Where M(X) represents the generated attention weight; represents broadcast addition, represents element-wise multiplication; N represents multiple satellite data sources.

[0081] Step 4: Use the fused remote sensing image to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area;

[0082] The flood identification in step 4 is a key step in flood warning and analysis. By using image processing techniques such as threshold segmentation, edge detection, and morphological processing, the boundaries and scope of the flood area can be effectively extracted. At the same time, other auxiliary information such as terrain, landform, and meteorological data can be combined to more accurately and comprehensively identify and analyze the flood area. The flood identification index NDWI is calculated as follows:

[0083]

[0084] NIR represents the reflectivity of the near infrared band, and SWIR represents the reflectivity of the short-wave infrared band. Deep learning is used to extract flood areas from optical remote sensing images. The input image is defined as I, and its corresponding feature map is defined as F. The goal is to generate a binary mask M, where M ij Indicates whether the pixel (i, j) belongs to the flood area; convolutional layers and pooling layers are used to extract high-level features of the image; the output feature map of the encoder is F enc ;

[0085] Decoder F enc Map it back to the original image size; use deconvolution layers and upsampling operations to restore the resolution; the output of the decoder is F dec ;

[0086] A convolutional layer is used to generate the water body segmentation probability map P, where P ij represents the probability that pixel (i, j) belongs to a water body; the probability map P is compared with a threshold τ = 0.7 to obtain the final binary mask M:

[0087]

[0088] The formula of the entire spinous process deep learning model for extracting flood area from optical remote sensing images can be expressed as:

[0089] M = Threshold(P, τ)

[0090] Threshold is a threshold processing function.

[0091] Step 5: Based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, flood warnings are carried out to obtain warning results;

[0092] In step 5, flood warning needs to comprehensively consider multiple factors, including historical flood data, meteorological, hydrological and other information; by analyzing and modeling these factors, the probability of flood occurrence, peak flow, arrival time and other information can be predicted, providing an important reference for disaster prevention and response; Flood warning model:

[0093] Flood prediction = Model flood (H, M, W)

[0094] Model flood is a flood prediction model based on Transformer architecture; H is a feature vector of historical flood data, including the frequency of past flood events, peak flow, flood duration, etc.; historical data can be obtained through statistical analysis; M is a feature value of meteorological factors, including rainfall, rainfall intensity, rainfall distribution, etc.; meteorological data can be obtained from meteorological stations, satellites or other meteorological sensors; W is a hydrological factor, including the topography, soil type, river network, water storage capacity, etc. of the basin; hydrological data can be obtained through hydrological models and geographic information systems;

[0095] The warning results in step 5 include: probability of flood occurrence, flood peak flow, and flood arrival time information.

[0096] Step 6: Conduct a comprehensive analysis of flood causes, flood processes, and flood impacts.

[0097] The comprehensive analysis described in step 6 is as follows:

[0098] Using geographic information system technology, flood data is overlaid and analyzed with geographic information such as topography, landforms, and population distribution to provide decision support for post-disaster rescue and reconstruction;

[0099] In step 6, flood warning requires analysis of the causes, processes, and impacts of floods, which can provide important experience and lessons for future disaster prevention and response. At the same time, it is also possible to use technical means such as geographic information systems (GIS) to overlay and analyze flood data with related geographic information, providing more scientific and effective decision-making support for post-disaster rescue and reconstruction.

[0100] The early warning system receives and processes various data in real time, including remote sensing images, meteorological information, hydrological data, etc., and uses advanced algorithms and models to analyze and predict floods. Once it detects that the flood risk exceeds the preset threshold, the system will immediately generate early warning information and promptly notify relevant departments and personnel through SMS, email, APP push, etc. At the same time, the early warning system can also provide real-time monitoring and dynamic updates of the flood range, helping decision makers to keep abreast of the latest flood conditions and changing trends and formulate scientific response measures.

[0101] like Figure 2 As shown, the system and method of the present invention extracts the flood area of ​​Canterbury in southeast England and infers the early warning flood area. By comparing with the flood area half a month later, the prediction accuracy of the system of the present invention can reach more than 80%.

[0102] The specific implementation of the system of the present invention is a flood monitoring and early warning system integrating multi-source remote sensing satellite data, including:

[0103] A data acquisition module for collecting various types of remote sensing satellite data on flooded areas;

[0104] The preprocessing module is used to perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data;

[0105] The feature fusion module is used to integrate the pre-processed remote sensing satellite data with different scale features to obtain better performance; the feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario;

[0106] The flood area recognition module is used to use the fused remote sensing images to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area;

[0107] The early warning module is used to conduct flood warnings based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, to obtain early warning results;

[0108] The comprehensive analysis module is used for comprehensive analysis of flood causes, flood processes, and flood impacts.

[0109] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0110] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.

Claims

1. A flood monitoring and early warning method based on multi-source remote sensing satellite data fusion, characterized in that: The following steps are involved: Step 1: Collect various types of remote sensing satellite data of the flooded area; Step 2: Perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data; Step 3: The preprocessed remote sensing satellite data are integrated with features of different scales to obtain better performance feature fusion; the feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario; Step 4: Use the fused remote sensing image to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area; Step 5: Based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, flood warning is carried out to obtain warning results; Step 6: Conduct a comprehensive analysis of flood causes, flood processes, and flood impacts.

2. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 1 is characterized in that: The remote sensing satellite data collection in step 1 needs to ensure the timeliness and accuracy of the data so as to obtain the latest surface information in a timely manner; at the same time, factors such as the coverage and resolution of different remote sensing satellite data need to be considered to ensure the comprehensiveness and reliability of the data.

3. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 2 is characterized in that: In step 2, data preprocessing can eliminate radiation errors in remote sensing images through radiation calibration and improve the radiation accuracy of images; geometric correction can eliminate geometric distortion in images and make images more consistent with the real surface morphology; atmospheric correction can eliminate the influence of atmosphere on remote sensing images and improve the clarity and contrast of images; distortion correction formula: r measured is the internal parameter of the satellite, measuring the radial distance; k1 and k2 are the radial distortion coefficients.

4. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 3 is characterized by: The data fusion in step 3 is to further improve the quality and resolution of remote sensing images; a revisit channel attention module is constructed for data fusion; The feature fusion based on the revisit channel attention module is extended from the same-layer scenario to the cross-layer scenario, including: short skip connections and long skip connections, and even internal feature integration; The revisit channel attention module includes a multi-scale channel fusion submodule, a local channel context aggregation submodule, and a global context fusion submodule; the input is the data feature map from different satellites, the channel and feature map of size H×W; these inputs are used for subsequent feature extraction and attention weight calculation; First, the multi-scale channel fusion submodule performs multi-scale feature fusion on the input multi-source satellite data. Given a source satellite data X, a channel and feature map of size H×W, the channel attention weight w is calculated as follows: w=σ(g(X))=σ(B(W2δ(B(W1(g(X)))))); where g(X) represents the global feature context, and is the global average pooling; δ represents the ReLU activation function, B represents batch normalization; σ is the Sigmoid function; this is achieved through a link with two fully connected layers, where W1 is the dimension reduction layer and W2 is the dimension increase layer; aggregating local and multi-source data context information.

5. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 4 is characterized by: The local channel context aggregation submodule realizes the aggregation of local context through the convolution operation PWConv; the key idea is that by changing the spatial pooling size, channel attention can be achieved at multiple scales; the local context needs to be added to the global context in the attention module; point-by-point convolution is selected as the local channel context aggregator, and the point-by-point channel interaction at each spatial position is utilized; the local channel context L(X) is calculated as follows: L(X)=B(PWConv1(δ(B(PWCon1(X))))) Among them, PWConv is a convolution with a convolution kernel of 1×1; L(X) retains and highlights the details in low-level features; Finally, the global context fusion submodule combines the local channel context with the global channel context to obtain the global feature representation X of multi-source data. g : Where M(X) represents the generated attention weight; represents broadcast addition, represents element-wise multiplication; N represents multiple satellite data sources.

6. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 5 is characterized by: The flood identification in step 4 is a key step in flood warning and analysis. By using image processing techniques such as threshold segmentation, edge detection, and morphological processing, the boundaries and scope of the flood area can be effectively extracted. At the same time, other auxiliary information such as terrain, landform, and meteorological data can be combined to more accurately and comprehensively identify and analyze the flood area. The flood identification index NDWI is calculated as follows: NIR represents the reflectivity of the near infrared band, and SWIR represents the reflectivity of the short-wave infrared band. Deep learning is used to extract flood areas from optical remote sensing images. The input image is defined as I, and its corresponding feature map is defined as F. The goal is to generate a binary mask M where M ij Indicates whether the pixel (i, j) belongs to the flood area; convolutional layers and pooling layers are used to extract high-level features of the image; the output feature map of the encoder is F enc ; Decoder F enc Map it back to the original image size; use deconvolution layers and upsampling operations to restore the resolution; the output of the decoder is F dec ; A convolutional layer is used to generate the water body segmentation probability map P, where P ij represents the probability that pixel (i, j) belongs to a water body; the probability map P is compared with a threshold τ to obtain the final binary mask M: The formula of the entire spinous process deep learning model for extracting flood area from optical remote sensing images can be expressed as: M=Threshold(P,τ) Threshold is a threshold processing function.

7. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 6 is characterized by: In step 5, flood warning needs to comprehensively consider multiple factors, including historical flood data, meteorological, hydrological and other information; by analyzing and modeling these factors, the probability of flood occurrence, peak flow, arrival time and other information can be predicted, providing an important reference for disaster prevention and response; Flood warning model: Flood prediction = Model flood (H,M,W) Model flood is a flood prediction model based on the Transformer architecture; H is a feature vector of historical flood data, including the frequency of past flood events, peak flow, flood duration, etc.; historical data can be obtained through statistical analysis; M is a feature value of meteorological factors, including rainfall, rainfall intensity, rainfall distribution, etc.; meteorological data can be obtained from meteorological stations, satellites or other meteorological sensors; W is a hydrological factor, including the topography, soil type, river network, water storage capacity, etc. of the basin; Hydrological data can be obtained through hydrological models and geographic information systems; The warning results in step 5 include: probability of flood occurrence, flood peak flow, and flood arrival time information.

8. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 7 is characterized by: The comprehensive analysis described in step 6 is as follows: Using geographic information system technology, flood data is overlaid and analyzed with geographic information such as topography, landform, and population distribution to provide decision support for post-disaster rescue and reconstruction.

9. The flood monitoring and early warning method based on multi-source remote sensing satellite data fusion according to claim 8 is characterized by: In step 6, flood warning requires analysis of the causes, processes, and impacts of floods, which can provide important experience and lessons for future disaster prevention and response. At the same time, geographic information system technology can be used to overlay and analyze flood data with related geographic information to provide more scientific and effective decision-making support for post-disaster rescue and reconstruction.

10. A flood monitoring and early warning system based on multi-source remote sensing satellite data fusion, characterized in that: include; A data acquisition module for collecting various types of remote sensing satellite data on flooded areas; The preprocessing module is used to perform radiometric calibration, geometric correction, and atmospheric correction preprocessing on the collected remote sensing satellite data to eliminate noise and interference in the data; The feature fusion module is used to integrate the pre-processed remote sensing satellite data with different scale features to obtain better performance; Extend the feature fusion based on the revisit channel attention module from the same-layer scenario to the cross-layer scenario; The flood area recognition module is used to use the fused remote sensing images to identify floods by using threshold segmentation, edge detection, and morphological processing to extract the boundaries and scope of the flood area; The early warning module is used to conduct flood warnings based on the identified flood areas and historical flood data, combined with meteorological and hydrological information, to obtain early warning results; The comprehensive analysis module is used for comprehensive analysis of flood causes, flood processes, and flood impacts.

Images