Flood diffusion simulation method and system for coupling convolutional neural network and cellular automaton

Abstract

The invention provides a flood diffusion simulation method and system coupled with a convolutional neural network and a cellular automaton. The method comprises the following steps: S1, acquiring an SAR image, a digital elevation model and a slope map of a research area; s2, gradually processing the SAR image of the research area based on a threshold segmentation method, an eight-connected area marking algorithm and a gray morphological algorithm to obtain a main body water body map of the research area; s3, inputting the main body water body map, the digital elevation model and the slope map of the research area into the trained CRU-Net model, and outputting a flood expansion potential map; and S4, updating the main body water body map of the research area by using the cellular automaton based on the flood expansion potential map in combination with the neighborhood constraint and the spatial constraint to obtain a main body water body map of the research area in the next time step, returning to the step S3 until a preset end condition is met, and outputting the main body water body map at the moment as a flood expansion range.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing emergency disaster prevention and control, specifically involving a flood diffusion simulation method and system that couples convolutional neural networks and cellular automata. Background Technology

[0002] The formation and evolution of floods are influenced by multiple factors, exhibiting high nonlinearity and complexity. Therefore, achieving accurate flood prediction and simulation remains a key technical challenge for hydrologists and technicians. While numerous hydrological and hydrodynamic physical models have been established to simulate surface runoff processes, the complexity of parameters and high learning costs continue to hinder the development of flood simulation. Traditional research largely relies on surface runoff hydrological parameters recorded by specialized instruments at hydrological stations, predicting and simulating flood flow, water depth, and other textual time-series data based on hydrodynamic equations. This approach lacks the intuitive representation of two-dimensional images and cannot describe the two-dimensional dynamic changes in flood diffusion. Furthermore, actual flood disasters present challenges such as the difficulty in acquiring multi-time-series satellite remote sensing data and the insufficient robustness and generalization ability of hydrodynamic models in flood prediction. Therefore, a data-driven flood diffusion simulation method is urgently needed. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, such as insufficient robustness and generalization ability of hydrodynamic models in flood prediction, and the difficulty in acquiring multi-time-series satellite remote sensing data during actual flood disasters, this invention provides a flood diffusion simulation method and system that couples convolutional neural networks and cellular automata. By using the CRU-Net model, the method transforms the identification of static ground features into capturing the dynamic physical diffusion process of floods, realizing an end-to-end, data-driven spatiotemporal flood prediction framework. Through the coupled convolutional neural network and cellular automata model, combined with neighborhood and spatial constraints, the method rapidly calculates the flood expansion range. Iterative computation solves the technical problem of the lack of a time dimension in a single convolutional neural network, achieving efficient, accurate, and scalable flood disaster prediction, and providing decision support for disaster emergency response.

[0004] According to one aspect of the present invention, a method for simulating flood diffusion by coupling a convolutional neural network and a cellular automata is provided, comprising:

[0005] S1. Acquire SAR images, digital elevation models, and slope maps of the study area;

[0006] S2. Based on threshold segmentation, eight-connected region labeling algorithm and grayscale morphology algorithm, the SAR image of the study area is processed step by step to obtain the main water body map of the study area.

[0007] S3. Input the main water body map, digital elevation model and slope map of the study area into the trained CRU-Net model, and output the flood expansion potential map.

[0008] S4. Combining neighborhood constraints and spatial constraints, based on the flood expansion potential map, use cellular automata to update the main water body map of the study area to obtain the main water body map of the study area in the next time step. Combining neighborhood constraints and spatial constraints, return to step S3 until the preset termination condition is met, and output the main water body map at this time as the flood expansion range.

[0009] As a further technical solution, the process of processing SAR images of the study area based on threshold segmentation and eight-connected region labeling algorithm includes:

[0010] Dual polarization calculations were performed on SAR images of the study area to obtain SDWI water index grayscale images. Water masks were extracted using the threshold segmentation method to obtain preliminary water maps.

[0011] The eight-connected region labeling algorithm is applied to traverse the preliminary water body map, calculate the pixel area of ​​each connected region in the map, and retain the connected region with the largest area as the main water body;

[0012] The main water body was processed using grayscale morphology to obtain the main water body image.

[0013] As a further technical solution, in the training dataset of the CRU-Net model, labels are set according to whether the main water body map changes in adjacent time steps. If the water body mask remains the same in adjacent time steps, it is determined that there has been no change, and the pixel label value is marked as 0; otherwise, the pixel label value is marked as 1.

[0014] As a further technical solution, the CRU-Net model includes an encoder, a decoder, intermediate layers, convolutional layers, and skip connections connecting the encoder and decoder. The encoder processes the input data of the CRU-Net model sequentially through several downsampling layers to extract encoded features at different scales. The intermediate layers are residual convolutional blocks that process the features extracted by the encoder's final downsampling layer to obtain deep features. The decoder includes several upsampling layers, corresponding one-to-one with the encoder's downsampling layers according to feature scale. The skip connections introduce the encoded features extracted by the upsampling layers into the corresponding downsampling layers, decoding the deep features layer by layer. The convolutional layers process the decoded features from the decoder's final upsampling layer and output a flood expansion potential map.

[0015] As a further technical solution, the downsampling layer sequentially includes a residual convolutional block, a convolutional attention module, and a max pooling.

[0016] As a further technical solution, the steps for updating the main water body map of the study area using a cellular automata model include:

[0017] The cellular automata model uses pixels in the main water body map as cells. It calculates the comprehensive probability of flood evolution for each cell by combining neighborhood constraints, boundary constraints, and flood expansion potential map to obtain a comprehensive probability map of flood evolution. The comprehensive probability map of flood evolution is then logically superimposed on the main water body map to obtain the main water body map of the study area in the next time step. The rule of logical superposition is: when the comprehensive probability of flood evolution of a cell exceeds a preset conversion threshold, the corresponding pixel in the main water body map is marked as a water body; otherwise, it is marked as a non-water body.

[0018] As a further technical solution, the formula for calculating the comprehensive probability of flood evolution is as follows:

[0019]

[0020] in, This represents the cellular flood expansion probability obtained from the flood expansion potential map; This represents the neighborhood constraint coefficient representing the proportion of water in a cell; This represents the boundary condition constraint coefficient of a cell.

[0021] According to another aspect of this specification, a flood diffusion simulation system coupled with a convolutional neural network and a cellular automata is provided, comprising:

[0022] The input module is used to acquire SAR images, digital elevation models, and slope maps of the study area.

[0023] The main water body extraction module is used to process the SAR image of the study area step by step based on the threshold segmentation method, the eight-connected region labeling algorithm and the gray-scale morphology algorithm to obtain the main water body map of the study area.

[0024] The flood expansion potential map generation module is used to input the main water body map, digital elevation model and slope map of the study area into the trained CRU-Net model and output the flood expansion potential map.

[0025] The main water body update simulation module is used to combine neighborhood constraints and spatial constraints, based on the flood expansion potential map, and use cellular automata to update the main water body map of the study area to obtain the main water body map of the study area in the next time step.

[0026] The output module is used to output the extent of flood expansion.

[0027] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to perform a flood diffusion simulation method coupled with a convolutional neural network and a cellular automata.

[0028] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a flood diffusion simulation method coupled with a convolutional neural network and a cellular automata.

[0029] Compared with existing technologies, the advantages of this invention are as follows: By mining the deep nonlinear features of the input data through the CRU-Net model, the potential laws governing the expansion of complex flood systems can be obtained without complex hydrodynamic equations; the label processing method is not the traditional static water body pixels, but rather the dynamic water body diffusion state; this feature allows the model learning objective of the CRU-Net model to shift from identifying static land features to capturing dynamic physical diffusion processes, realizing an end-to-end, data-driven flood spatiotemporal prediction framework; the neighborhood constraints and boundary condition constraints in the cellular automata transformation rules more realistically fit the complex geographical changes, while the iterative operation of the cellular automata adds a time dimension to the deep learning model; during flood disasters, only one phase of high-precision water body extraction data is needed to complete the dynamic simulation of the flood prediction range, effectively solving the technical problem of not being able to use multi-time series data for model calculation during floods. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart illustrating a flood diffusion simulation method coupled with a convolutional neural network and a cellular automata, provided for an embodiment of the present invention;

[0032] Figure 2 This is a high-precision main water body map of the sample area in this embodiment of the invention;

[0033] Figure 3 This is a schematic diagram of the CRU-Net model in an embodiment of the present invention;

[0034] Figure 4 This is a schematic diagram of the output results of the CRU-Net model in an embodiment of the present invention;

[0035] Figure 5 This is a schematic diagram showing the comparison between the flood inundation range and the simulated flood range of the sample area in an embodiment of the present invention;

[0036] Figure 6 This is a graph showing the change in the accuracy of the model's prediction of the flood range in an embodiment of the present invention;

[0037] Figure 7 This is a graph showing the change in the recall rate of the flood range predicted by the model in an embodiment of the present invention;

[0038] Figure 8 This is a graph showing the change in the F1 score of the flood range predicted by the model in an embodiment of the present invention;

[0039] Figure 9 This is a graph showing the variation of the Kappa coefficient in the flood range predicted by the model in this embodiment of the invention.

[0040] Figure 10 A schematic diagram of a flood diffusion simulation system coupled with a convolutional neural network and a cellular automata provided in an embodiment of the present invention;

[0041] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0042] It should be noted that:

[0043] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0044] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0046] like Figure 1 As shown, a flood diffusion simulation method coupling convolutional neural networks and cellular automata includes:

[0047] Step 1: Acquire SAR images, digital elevation models, and slope maps of the study area;

[0048] Step 2: Based on the threshold segmentation method, the eight-connected region labeling algorithm and the gray-scale morphology algorithm, the SAR image of the study area is processed step by step to obtain the main water body map of the study area;

[0049] Step 3: Input the main water body map, digital elevation model and slope map of the study area into the trained CRU-Net model, and output the flood expansion potential map;

[0050] Step 4: Combining neighborhood constraints and spatial constraints, based on the flood expansion potential map, use cellular automata to update the main water body map of the study area to obtain the main water body map of the study area in the next time step. Combining neighborhood constraints and spatial constraints, return to step S3 until the preset termination condition is met, and output the main water body map at this time as the flood expansion range.

[0051] In step 1, the SAR image of the study area is obtained by preprocessing radar remote sensing images. The preprocessing operations of Sentinel-1 and GF-3 series images mainly include orbit correction, thermal noise removal, radiometric calibration, filtering, terrain correction, decibel reduction, image mosaicking and cropping, and resampling.

[0052] Step 2, the process of processing SAR images of the study area based on the threshold segmentation method and the eight-connected region labeling algorithm, includes:

[0053] Step 2-1: Perform dual-polarization calculation on the radar image of each preprocessed study area to obtain the SDWI water index grayscale image, and use the threshold segmentation method to extract the water mask to obtain the preliminary water map.

[0054] Step 2-2: Apply the eight-connected region labeling algorithm to traverse the preliminary water body map, calculate the pixel area of ​​each connected region in the map, and retain the connected region with the largest area as the main water body;

[0055] Steps 2-3: Then, grayscale morphology is used to process the main water body to obtain the main water body image.

[0056] Specifically, in step 2-1, by utilizing the difference in backscattering of water bodies under different polarization modes, the SDWI index calculation can significantly enhance the feature contrast between water bodies and non-water bodies (such as vegetation, soil, and buildings), suppress background interference, and provide a highly discriminative SDWI water body index grayscale image for subsequent threshold segmentation. Histogram statistical analysis is performed on the generated SDWI water body index grayscale image. Through experimental comparison, the optimal segmentation threshold is selected. Pixels larger than the threshold are marked as water body pixels, and the rest are marked as background, thereby generating a binary mask for preliminary water body extraction. This achieves rapid separation of water body targets from the background, completes pixel-level preliminary classification, delineates all potential water body regions, and obtains a preliminary water body map.

[0057] Preferably, in step 2-2, the 8-connected algorithm is used to identify different water bodies in the image, and the water bodies are sorted according to the number of pixels contained in each water body, and the largest water body is selected as the main water body.

[0058] Specifically, in steps 2-3, grayscale morphological processing is performed on the main water body image. This includes using a closing operation (dilation followed by erosion) to fill small voids within the water body area, repairing "fractures" or "voids" within the water body caused by uneven reflection of radar signals or partial obstruction, thereby removing water noise and filling voids. Preferably, a 7×7 rectangular structural element is used for dilation and erosion operations. In this embodiment, the high-precision water body extraction image of the sample area is shown below. Figure 2 As shown.

[0059] Step 3, specifically the training process of the CRU-Net model, includes:

[0060] Step 3-1: Construct a training dataset based on the SAR image set, digital elevation model, and slope map of the study area;

[0061] Step 3-2: Construct the CRU-Net model; the CRU-Net model processes the main water body map, digital elevation model and slope map of the study area, and finally outputs the flood expansion potential map;

[0062] Step 3-3: Train the CRU-Net model using the training dataset and output the trained model.

[0063] In step 3, the CRU-Net model's training dataset is labeled based on whether the main water body map changes in adjacent time steps. If a pixel remains consistent in adjacent time steps, it is determined that it has not changed and the pixel's label value is marked as 0; otherwise, the pixel's label value is marked as 1.

[0064] The label processing method is not the traditional static water body pixels, but the dynamic water body diffusion state; this feature enables the model learning objective to change from identifying static land features to capturing dynamic physical diffusion processes, realizing an end-to-end, data-driven flood spatiotemporal prediction framework.

[0065] Specifically, in step 3-1, the training dataset is constructed as follows: high-precision main water body data, DEM and slope data are sliced, the slice size is 96×96, the overlap is set to 50%, the continuity is maintained and the problem of computer memory or video memory overflow is solved.

[0066] The steps for acquiring high-precision main water body data include: acquiring a SAR image set of the study area, processing it according to step 2, and obtaining high-precision main water body data.

[0067] Specifically, the acquisition process of SAR image set, DEM, and slope data for the study area is as follows: Radar images have the advantage of penetrating cloud layers and being unaffected by weather conditions. Based on visual inspection methods, all radar images within the study area are selected to acquire Sentinel-1 and GF-3 series satellite images. These images include both pre-flood and post-flood images, and maintain consistent imaging modes, polarization methods, and orbital parameters. Sentinel-1 and GF-3 remote sensing images from multiple acquisition days are sequentially processed through orbit correction, thermal noise removal, radiometric calibration, filtering, terrain correction, decibel reduction, mosaicking and cropping, and resampling to obtain preprocessed SAR images of the study area. Precise orbital ephemeris data is then used to... Ephemerides updates the orbital state vector of the image, corrects satellite position and velocity information, improves the geometric positioning accuracy of the image, eliminates additive thermal noise generated by the radar receiver system itself to improve the signal-to-noise ratio of the image, reduces interference to low-scattering areas (such as calm water bodies), converts the pixel digital quantization values ​​(DN values) of the image into physically meaningful radar backscattering coefficients, eliminates the influence of sensor instability, and makes images from different time phases comparable. For the speckle noise unique to SAR images, the En-Lee filtering algorithm is used for denoising, suppressing noise while preserving edge texture information, and an external digital elevation model (DE) is introduced. The imagery was re-projected onto the WGS84 geographic coordinate system to eliminate geometric distortions caused by topographic relief. The backscattering coefficients after radiometric calibration were logarithmically transformed and converted to decibels (dB) to enhance the visualization and contrast of the data. The resulting pre-processed SAR image set for the study area was obtained. ArcGIS raster mosaicking and clipping tools were used to stitch together data from the same date. Vector range data was used for clipping, and resampling tools were used to unify the image resolution to 30m to maintain the consistency of data size. Digital elevation model (DEM) data for the study area was acquired, and slope images were obtained using ArcGIS slope calculation tools.

[0068] In step 3-2, such as Figure 3As shown, the CRU-Net model has the same encoder-decoder architecture as the U-Net model. The encoder merges features with the decoder through skip connections. During downsampling, the feature map's channel count is changed by passing through a residual convolutional block (Res Block) and a CBAM (Convolution Block Attention Module) module to extract channel and spatial features. Max pooling is then used to further extract features at different scales. During upsampling, deconvolution is used to restore the feature map's size, and a feature skip connection is made with the corresponding size feature map from the downsampling process. After the connection is completed, residual convolutional blocks and convolutional attention modules are used to further extract features. Finally, a flood expansion potential map with one channel and the same size as the input data is output through the convolutional block.

[0069] Specifically, the CRU-Net model includes an encoder, a decoder, intermediate layers, convolutional layers, and skip connections connecting the encoder and decoder. The encoder processes the input data of the CRU-Net model sequentially through several downsampling layers to extract encoded features at different scales. The intermediate layers are residual convolutional blocks that process the features extracted by the encoder's final downsampling layer to obtain deep features. The decoder includes several decoding layers, each corresponding to a downsampling layer of the encoder in terms of feature scale. The skip connections introduce the encoded features extracted by the downsampling layers into the corresponding downsampling layers, decoding the deep features layer by layer. The convolutional layers process the decoded features of the decoder's final upsampling layer, outputting a flood expansion potential map.

[0070] First, the encoder (downsampling process) processes the input feature map, going through several downsampling stages (one downsampling layer), extracting hierarchical features from the input feature map. The intermediate layer receives the features from the last downsampling stage and processes them to obtain the features to be decoded.

[0071] Each downsampling layer consists of a residual convolutional block (Res Block), a convolutional attention module (CBAM Block), and a max pooling operation. At the end of the downsampling stage, max pooling is used to perform downsampling operations, gradually reducing the feature map size and expanding the receptive field.

[0072] The intermediate layer is located between the encoder and the decoder. It receives the feature map output by the final downsampling layer and uses a residual convolutional block for deep feature extraction. This layer acts as the "bottleneck" of the encoder network and gathers the highest level of semantic information.

[0073] Furthermore, the residual convolutional block sequentially contains a 1×1 convolutional layer, a 3×3 convolutional layer, and a Leaky ReLU activation function. Each layer is connected through a batch normalization layer, introducing residual links, which effectively alleviates the gradient vanishing problem in deep networks, accelerates model convergence, and extracts deep semantic features.

[0074] The Convolutional Attention Module (CBAM) is a concatenated module consisting of a Channel Attention Module and a Spatial Attention Module. The Channel Attention Module automatically identifies which channels are more critical (e.g., polarization channels with strong water reflection) and suppresses irrelevant channels to focus on key features (e.g., water texture). The Spatial Attention Module focuses on "where" the key features are located (e.g., water boundaries). The combination of the two allows the model to adaptively weight the feature maps.

[0075] The Channel Attention Module processes the feature map of the input CBAM module, performs max pooling and average pooling on the feature map, and feeds the two pooling results into a shared multilayer perceptron (MLP) to learn the dependencies between channels and output channel attention weights to represent the importance of each channel. The feature map of the input CBAM module is weighted by the channel attention weights to obtain the (Channel-refined Feature).

[0076] The Spatial Attention Module processes the channel-refined features by performing max pooling and average pooling on them to generate two spatial response maps. The spatial response maps are then concatenated and passed through a convolutional layer to generate a spatial attention map. The spatial attention weights are obtained by passing the sigmoid function, and the channel-refined features are then weighted by spatial attention to obtain the refined features.

[0077] Secondly, the decoder (upsampling process) is used to restore the spatial resolution of the feature map. Each upsampling stage (decoding layer) includes: deconvolution operation (up sampling), which uses transposed convolution to upsample the feature map and gradually restore the size of the feature map.

[0078] Third, the feature skip connection concatenates the upsampled feature map with the corresponding feature map in the encoder, compensating for the spatial details lost during downsampling and fusing shallow localization information with deep semantic information to improve the precision of water body edge segmentation. After feature fusion and recovery, the hybrid features are passed through the residual convolutional block (Res Block) and the convolutional attention module (CBAM) again to further refine and extract the recovered features.

[0079] Fourth, the output layer (Out Block) uses a convolutional layer. After multiple layers of decoding, a final convolutional layer (Conv 1x1) maps the multi-channel features to an output with one channel. The output size is consistent with the input data, generating a flood expansion potential map (or water body prediction probability map). The value of each pixel in the map represents the probability that the point belongs to a water body or that flood expansion will occur.

[0080] As an optional implementation method, such as Figure 3 As shown, the CRU-Net model encoder includes a first upsampling layer, a second upsampling layer, a third upsampling layer, a fourth upsampling layer, and a convolutional layer, which are sequentially connected to residual convolutional blocks and CBAM modules. The encoder and decoder are connected through an intermediate layer, which is a residual convolutional block. The first downsampling layer and the fourth upsampling layer are connected in a skip connection, the second downsampling layer and the third upsampling layer are connected in a skip connection, the third downsampling layer and the second upsampling layer are connected in a skip connection, and the fourth downsampling layer and the first upsampling layer are connected in a skip connection.

[0081] The first downsampling stage receives preprocessed 3-channel input data (the data in each channel are the main water body map, DEM, and slope of the current time phase, respectively). After processing by a residual convolution block with a convolution window size of 3×3 and edge padding, the number of feature map channels is expanded from 3 to 8, and the size after downsampling becomes (S / 2). 2 The second downsampling layer receives the 8-channel feature map output from the first layer. After processing by a residual convolution block with a convolution window size of 3×3 and edge padding, the number of feature map channels is expanded from 8 to 16, and the size after downsampling becomes (S / 4). 2 The third downsampling layer receives the 16-channel feature map output from the second layer. After processing by a residual convolution block with a convolution window size of 3×3 and edge padding, the number of feature map channels is expanded from 16 to 32, and the size becomes (S / 8) after downsampling. 2 The fourth downsampling layer receives the 32-channel feature map output from the third layer. After processing by a residual convolution block with a convolution window size of 3×3 and edge padding, the number of feature map channels is expanded from 32 to 64, and the size becomes (S / 16) after downsampling. 2The intermediate layer, located between the encoder and decoder, receives the 64-channel feature map output from the fourth downsampling layer. It then uses residual convolutional blocks with 3×3 convolutional windows and edge padding for deep feature extraction, outputting a feature map with 128 channels while maintaining its original size. This layer acts as the "bottleneck" of the encoder network, converging the highest-level semantic information.

[0082] The intermediate layer receives a feature map with 64 channels and uses a residual convolution block with a convolution window size of 3×3 and edge padding to output a feature map with 128 channels and unchanged size.

[0083] The convolutional layer receives input data with 8 channels and, using a 3×3 convolutional window with edge padding, outputs a flood expansion potential map with 1 channel. Figure 2 Using the combined sample area DEM and slope map as input, the final output is a flood expansion potential map, as shown below. Figure 4 As shown.

[0084] In step 3-3, the CRU-Net model is trained using the slice data and corresponding labels in the training dataset constructed in step 3-1. Forward propagation inputs the preprocessed multi-channel slice data into the CRU-Net model, which, after encoding, decoding, and attention weighting, outputs the prediction result. The flood expansion potential map output by the model is compared with the true label data (GroundTruth), and the loss function value (Loss) is calculated. Based on the calculated loss value, the gradient is calculated using the backpropagation algorithm, and the Adam optimizer is used to iteratively update the weights and bias parameters in the network. This process is repeated until the loss value converges or reaches the preset number of iterations (Epochs). The optimal model parameters are then saved, completing the model training part.

[0085] Step 4, which involves updating the main water body map of the study area using a cellular automata model, includes the following steps:

[0086] The cellular automata model uses pixels in the main water body map as cells. It calculates the comprehensive probability of flood evolution for each cell by combining neighborhood constraints, boundary constraints, and flood expansion potential map to obtain a comprehensive probability map of flood evolution. The comprehensive probability map of flood evolution is then logically superimposed on the main water body map to obtain the main water body map of the study area in the next time step. The rule of logical superposition is: when the comprehensive probability of flood evolution of a cell exceeds a preset conversion threshold, the corresponding pixel in the main water body map is marked as a water body; otherwise, it is marked as a non-water body.

[0087] Among them, the neighborhood constraint is represented by the probability of water body proportion in the cell, indicating the proportion of water body pixels within the preset neighborhood. Optionally, the neighborhood range can be selected as a 31×31 neighborhood. The boundary condition constraint is set according to the effective range mask or extreme terrain limitation of the study area, and is represented by the boundary constraint coefficient. The cell boundary constraint coefficient is 1 in the effective area and 0 in the invalid area or outside the boundary. The added neighborhood constraint and boundary condition constraint more realistically fit the changing process of complex geography.

[0088] The formula for calculating the comprehensive probability of flood evolution is shown below:

[0089]

[0090] in, This represents the cellular flood expansion probability obtained from the flood expansion potential map; This represents the neighborhood constraint coefficient representing the proportion of water in a cell; This represents the boundary condition constraint coefficient of a cell.

[0091] To ensure that the flood expansion simulation conforms to the geographic spatial neighborhood patterns and physical boundary constraints, the above formula is combined to construct a comprehensive transformation probability formula. This formula is then combined with the local transformation rules of cellular automata to perform spatiotemporal dynamic simulation. For each cell (pixel) within the study area, three components are sequentially obtained: the cell flood expansion probability, the cell water body proportion neighborhood constraint coefficient, and the cell boundary condition constraint coefficient. These components are then multiplied at the pixel level to characterize the flood evolution probability.

[0092] Step 4 essentially involves multiplying the flood expansion potential map, the cellular water body proportion neighborhood constraint, and the boundary condition constraint sequentially, and then overlaying them onto the initial real water body map. Based on the threshold, the expansion range of the flood is simulated. The generated data is then used as input data for the model training and prediction steps to generate a new flood expansion potential map. The flood expansion range at the next moment is simulated through the cellular automata prediction step. This process continues iteratively until the flood no longer shows significant expansion.

[0093] In a specific implementation, the probability map is combined with the actual water body map at the current time step (the main water body map output by the CRU-Net model). Figure 4 The conversion threshold T for logical superposition is set to 0.1. When the flood evolution comprehensive probability of a certain cell is... If the state of the corresponding pixel in the cell is determined by the condition, it is marked as water in the next time step; otherwise, it is marked as non-water. This process generates... The results of comparing the simulated flood extent at each time step with the flood inundation extent of the sample area and the simulated flood extent are as follows: Figure 5 As shown.

[0094] The above generated Time-step flood extent map, as The real water body map at each time step is recombined with static data (DEM, slope) to construct new input slice data; the new input data is then loaded back into the CRU-Net model for prediction, generating... The flood expansion potential map is generated at each time step, and the logical overlay step is repeated. The number of iterations is set to 100 to ensure that the flood does not expand significantly. The final output image is the maximum inundation range of the simulated flood. During the iteration process, various indicators are as follows: Figures 6-9 As shown, Figure 6 This is a graph showing the change in accuracy. Figure 7 A graph showing the change in recall rate; Figure 8 The graph shows the changes in F1 scores; Figure 9 The Kappa coefficient variation graph shows that, in the event of a flood disaster, this invention enables the simulation of the flood expansion process using only one phase of water body image, solving the problem of water diffusion simulation under conditions of missing multi-temporal water body data; and it solves the technical problem of the lack of time dimension of a single convolutional neural network by iterative operation of a coupled model of convolutional neural network and cellular automata.

[0095] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a flood diffusion simulation system coupled with a convolutional neural network and a cellular automata. This system is used to execute a flood diffusion simulation method coupled with a convolutional neural network and a cellular automata from the above method embodiments.

[0096] See Figure 10 The system includes:

[0097] The system comprises the following modules: an input module for acquiring SAR imagery, digital elevation model (DEM), and slope map of the study area; a main water body extraction module for progressively processing the SAR imagery of the study area using threshold segmentation, an eight-connected region labeling algorithm, and grayscale morphology algorithm to obtain the main water body map of the study area; a flood expansion potential map generation module for inputting the main water body map, DEM, and slope map of the study area into the trained CRU-Net model and outputting the flood expansion potential map; a main water body update simulation module for updating the main water body map of the study area using cellular automata based on the flood expansion potential map, combining neighborhood constraints and spatial constraints, to obtain the main water body map of the study area for the next time step; and an output module for outputting the flood expansion range.

[0098] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0099] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 11 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0100] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0101] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0102] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0103] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0106] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute a flood diffusion simulation method coupled with a convolutional neural network and a cellular automata.

[0107] In summary, this invention provides a flood diffusion simulation method coupled with convolutional neural networks and cellular automata, belonging to the field of remote sensing emergency disaster prevention and control. The method includes the following steps: preparing high-precision water body data, DEM, and slope data of the flooded area after radar image processing; slicing the data as model input; performing downsampling operations through ResBlock and CBAM convolutional block attention module to extract channel features and spatial features of feature maps at each scale; during upsampling, using deconvolution operation to restore the feature map size and stitching it with the feature map of the same size during downsampling to finally obtain a flood diffusion potential map of the same size as the input data; using the output diffusion potential map, neighborhood constraints, and boundary constraints as the transformation rules of the cellular automata to output the flood simulation image at the next time step; the simulation results can continue to be used as model input for iteration to finally obtain the flood diffusion range.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A flood diffusion simulation method coupled with a convolutional neural network and a cellular automata, characterized in that, include: S1. Acquire SAR images, digital elevation models, and slope maps of the study area; S2. Based on threshold segmentation, eight-connected region labeling algorithm and grayscale morphology algorithm, the SAR image of the study area is processed step by step to obtain the main water body map of the study area. S3. Input the main water body map, digital elevation model and slope map of the study area into the trained CRU-Net model, and output the flood expansion potential map. S4. Combining neighborhood constraints and spatial constraints, based on the flood expansion potential map, use cellular automata to update the main water body map of the study area, obtain the main water body map of the study area in the next time step, return to step S3, until the preset termination condition is met, and output the main water body map at this time as the flood expansion range.

2. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 1, characterized in that, The process of processing SAR images of the study area based on threshold segmentation and eight-connected region labeling algorithm includes: Dual polarization calculations were performed on SAR images of the study area to obtain SDWI water index grayscale images. Water masks were extracted using the threshold segmentation method to obtain preliminary water maps. The eight-connected region labeling algorithm is applied to traverse the preliminary water body map, calculate the pixel area of ​​each connected region in the map, and retain the connected region with the largest area as the main water body; The main water body was processed using grayscale morphology to obtain the main water body image.

3. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 1, characterized in that, In the training dataset of the CRU-Net model, labels are set according to whether the main water body map changes in adjacent time steps. If the water body mask remains consistent in adjacent time steps, it is determined that there has been no change, and the pixel label value is marked as 0; otherwise, the pixel label value is marked as 1.

4. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 1, characterized in that, The CRU-Net model comprises an encoder, a decoder, intermediate layers, convolutional layers, and skip connections connecting the encoder and decoder. The encoder processes the input data of the CRU-Net model sequentially through several downsampling layers, extracting encoded features at different scales. The intermediate layers are residual convolutional blocks that process the features extracted by the encoder's final downsampling layer to obtain deep features. The decoder comprises several upsampling layers, corresponding one-to-one with the encoder's downsampling layers according to feature scale. The skip connections introduce the encoded features extracted by the upsampling layers into the corresponding downsampling layers, decoding the deep features layer by layer. The convolutional layers process the decoded features from the decoder's final upsampling layer, outputting a flood expansion potential map.

5. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 4, characterized in that, The downsampling layer consists of a residual convolutional block, a convolutional attention module, and a max pooling layer.

6. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 1, characterized in that, The steps for updating the main water body map of the study area using a cellular automata model include: The cellular automata model uses pixels in the main water body map as cells. It calculates the comprehensive probability of flood evolution for each cell by combining neighborhood constraints, boundary constraints, and flood expansion potential map to obtain a comprehensive probability map of flood evolution. The comprehensive probability map of flood evolution is then logically superimposed on the main water body map to obtain the main water body map of the study area in the next time step. The rule of logical superposition is: when the comprehensive probability of flood evolution of a cell exceeds a preset conversion threshold, the corresponding pixel in the main water body map is marked as a water body; otherwise, it is marked as a non-water body.

7. The flood diffusion simulation method coupled with a convolutional neural network and a cellular automata as described in claim 6, characterized in that, The formula for calculating the comprehensive probability of flood evolution is shown below: ； in, This represents the cellular flood expansion probability obtained from the flood expansion potential map; This represents the neighborhood constraint coefficient representing the proportion of water in a cell; This represents the boundary condition constraint coefficient of a cell.

8. A flood diffusion simulation system coupled with a convolutional neural network and a cellular automata, characterized in that, include: The input module is used to acquire SAR images, digital elevation models, and slope maps of the study area; The main water body extraction module is used to process the SAR image of the study area step by step based on the threshold segmentation method, the eight-connected region labeling algorithm and the gray-scale morphology algorithm to obtain the main water body map of the study area. The flood expansion potential map generation module is used to input the main water body map, digital elevation model and slope map of the study area into the trained CRU-Net model and output the flood expansion potential map. The main water body update simulation module is used to combine neighborhood constraints and spatial constraints, based on the flood expansion potential map, and use cellular automata to update the main water body map of the study area to obtain the main water body map of the study area in the next time step. The output module is used to output the extent of flood expansion.

9. An electronic device, characterized in that, The method includes a memory and a processor, the memory storing program instructions that are executed by the processor, the processor invoking the program instructions to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1 to 7.

Images