Flood monitoring method and device, electronic equipment and computer readable storage medium

Abstract

The application provides a flood monitoring method and device, electronic equipment and computer readable storage medium, wherein the method comprises: acquiring digital elevation data, multispectral image data and time series Sentinel-1 data of a flood occurrence area; determining terrain features according to the digital elevation data; selecting a plurality of water sample areas and non-water sample areas from the multispectral image data as training samples; for each Sentinel-1 data, performing image segmentation on the Sentinel-1 data by using a multi-resolution segmentation algorithm to obtain an image segmentation result; extracting polarization band features, texture features, water body index features and time SAR features of the Sentinel-1 data, and taking each feature as an input of an initial random forest model; training the initial random forest model by using the training samples, so that the trained random forest model has the ability to distinguish water body areas and non-water body areas in the Sentinel-1 data. By the method, the monitoring accuracy of the flood is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flood monitoring, and in particular to a flood monitoring method and device, an electronic device, and a computer readable storage medium. BACKGROUND

[0002] Floods are natural disasters caused by heavy rain, snowmelt, dam breaches, and overflow of lakes or rivers. They cause significant damage to the natural environment and loss of life every year on a global scale. With climate change leading to an increase in extreme precipitation events, it is expected that the frequency of future flood events will further increase, thereby affecting people's livelihoods. Therefore, effective flood disaster monitoring is crucial to minimize loss of life, property, and infrastructure, as well as industrial, agricultural, and other losses. SUMMARY

[0003] Therefore, the present application aims to provide a flood monitoring method, device, electronic device, and computer readable storage medium to improve the monitoring accuracy of floods.

[0004] In a first aspect, the embodiments of the present application provide a flood monitoring method, comprising:

[0005] obtaining digital elevation data, multispectral image data, and time series Sentinel-1 data in a monitoring time period of a flood-occurring area; wherein the monitoring time period is the time period from before the flood disaster to after the flood disaster in the flood-occurring area;

[0006] determining the topographic features of the flood-occurring area according to the digital elevation data, and selecting a plurality of water body sample areas and a plurality of non-water body sample areas from the multispectral image data as training samples;

[0007] for each Sentinel-1 data in the time series Sentinel-1 data, performing image segmentation on the Sentinel-1 data using a multi-resolution segmentation algorithm to obtain a corresponding image segmentation result; wherein the image segmentation result contains the segmentation image of each ground feature in the Sentinel-1 data;

[0008] extracting the polarization band feature, texture feature, water body index feature, and time SAR feature of the Sentinel-1 data;

[0009] The terrain feature, the image segmentation result of the Sentinel-1 data, the polarization band feature, the texture feature, the water body index feature, and the time SAR feature are taken as inputs of an initial random forest model corresponding to the Sentinel-1 data, and the initial random forest model is trained by using the training sample, so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish the water body region from the non-water body region in the Sentinel-1 data.

[0010] With reference to the first aspect, in a first possible implementation manner of the first aspect, before the image segmentation of the Sentinel-1 data is performed by using the multi-resolution segmentation algorithm to obtain the corresponding image segmentation result, the method further includes:

[0011] The Sentinel-1 data is preprocessed to obtain preprocessed Sentinel-1 data, and the preprocessing includes applying an orbit file, removing GRD boundary noise, removing thermal noise, radiometric calibration, filtering processing, terrain correction, and converting backscattering intensity into decibels.

[0012] The image segmentation of the Sentinel-1 data is performed by using the multi-resolution segmentation algorithm to obtain the corresponding image segmentation result, and the image segmentation includes:

[0013] An optimal segmentation scale estimation tool is called to determine the optimal segmentation scale parameter, the optimal shape factor value, and the compactness parameter of the preprocessed Sentinel-1 data.

[0014] The determined optimal segmentation scale parameter, the optimal shape factor value, and the compactness parameter are input into the multi-resolution segmentation algorithm, and the image segmentation of the preprocessed Sentinel-1 data is performed by using the multi-resolution segmentation algorithm to obtain the image segmentation result.

[0015] With reference to the first aspect, in a second possible implementation manner of the first aspect, the polarization band feature, the texture feature, the water body index feature, and the time SAR feature of the Sentinel-1 data are extracted, and the extraction includes:

[0016] The polarization ratio, the polarization difference, the polarization sum, and the polarization product are calculated based on the VV polarization and the VH polarization of the Sentinel-1 data to obtain the polarization band feature of the Sentinel-1 data.

[0017] The gray level co-occurrence matrix is used to extract SAR dual-polarization texture features of the VV polarization and the VH polarization of the Sentinel-1 data, to obtain texture features of the Sentinel-1 data; the texture features are used to reflect the difference of ground objects in space;

[0018] Based on the VV polarization and the VH polarization of the Sentinel-1 data, a dual-polarization water body index is calculated, and the dual-polarization water body index is used as a water body index feature of the Sentinel-1 data; the water body index feature is used to highlight the water body, so as to distinguish the water body from other ground objects;

[0019] According to the backscattering data of the Sentinel-1 data and the time series Sentinel-1 data, a time SAR feature of the Sentinel-1 data is calculated; the time SAR feature includes a backscattering anomaly and a Z-score; the backscattering anomaly is used to represent the difference between the backscattering during the flood disaster and the average backscattering during the non-flood period; the Z-score is used to represent the difference between the backscattering during the flood disaster and the average backscattering of the monitoring period.

[0020] With reference to the first aspect, embodiments of the present application provide a third possible implementation manner of the first aspect, wherein the terrain features include elevation, slope, slope direction, and mountain shadow.

[0021] With reference to the first aspect, embodiments of the present application provide a fourth possible implementation manner of the first aspect, wherein the terrain features, the image segmentation result of the Sentinel-1 data, the polarization band features, the texture features, the water body index features, and the time SAR features are used as inputs of an initial random forest model corresponding to the Sentinel-1 data, and the initial random forest model is trained by using the training samples, including:

[0022] The terrain features, the polarization band features, the texture features, the water body index features, and the time SAR features of the Sentinel-1 data are input into the initial random forest model corresponding to the Sentinel-1 data, and the importance of each feature is determined by the initial random forest model;

[0023] According to the importance of each feature, a feature is sequentially added to the initial random forest model from high to low, to determine the classification accuracy of the initial random forest model after adding each feature;

[0024] According to the change of the classification accuracy of the initial random forest model, the features added in the initial random forest model when the classification accuracy is the highest are determined as the target features corresponding to the Sentinel-1 data;

[0025] The image segmentation result corresponding to the Sentinel-1 data and each target feature are taken as the input of an initial random forest model corresponding to the Sentinel-1 data, and the initial random forest model is trained by using the training sample.

[0026] With reference to the fourth possible implementation manner of the first aspect, the fourth possible implementation manner of the first aspect is provided in the embodiments of the present application, and the method further includes:

[0027] For each Sentinel-1 data, after obtaining the trained random forest model corresponding to the Sentinel-1 data, the feature values of each target feature corresponding to the Sentinel-1 data are fused into the Sentinel-1 data to obtain fused Sentinel-1 data;

[0028] The fused Sentinel-1 data and the image segmentation result corresponding to the Sentinel-1 data are input into the random forest model, and the water body region and the non-water body region in the fused Sentinel-1 data are identified based on the random forest model.

[0029] With reference to the fifth possible implementation manner of the first aspect, the sixth possible implementation manner of the first aspect is provided in the embodiments of the present application, and the method further includes:

[0030] Based on the identified water body region and non-water body region in the fused Sentinel-1 data, a binary classification image for distinguishing the water body region and the non-water body region is generated.

[0031] After obtaining the binary classification image corresponding to each Sentinel-1 data, for each pixel in the binary classification image, the frequency of the pixel being identified as a water body is calculated, and the frequency is used to reflect the duration of the water body at the pixel in the monitoring time period.

[0032] A plurality of pixels are selected from all the pixels as sample pixels, and each sample pixel is determined to be a permanent water body pixel or a flood pixel in a real situation.

[0033] For each preset frequency threshold, compare the respective frequency of each of the sample pixels with the preset frequency threshold to infer whether each of the sample pixels is a permanent water body pixel or a flood pixel based on the preset frequency threshold;

[0034] According to the inferred permanent water body pixels or flood pixels and the determined permanent water body pixels or flood pixels in the real situation, calculate the accuracy of the preset frequency threshold in distinguishing permanent water body pixels and flood pixels;

[0035] According to the respective accuracy of each preset frequency threshold, select the preset frequency threshold with the highest accuracy as a target preset frequency threshold;

[0036] For each of the pixels of the two-class image, compare the frequency of the pixel with the target preset frequency threshold, if the frequency of the pixel is greater than the target preset frequency threshold, it indicates that the pixel is a permanent water body pixel; if the frequency of the pixel is less than or equal to the target preset frequency threshold, it indicates that the pixel is a flood pixel.

[0037] In a second aspect, the embodiments of the present application also provide a flood monitoring device, comprising:

[0038] An acquisition module is configured to acquire digital elevation data of a flood-occurred area, multispectral image data, and time series Sentinel-1 data in a monitoring time period; wherein the monitoring time period is a time period from before the flood-occurred area occurs a flood disaster to after the flood-occurred area occurs a flood disaster;

[0039] A determination module is configured to determine a topographic feature of the flood-occurred area according to the digital elevation data, and select a plurality of water body sample areas and a plurality of non-water body sample areas from the multispectral image data as training samples;

[0040] A segmentation module is configured to, for each of the time series Sentinel-1 data, perform image segmentation on the Sentinel-1 data by using a multi-resolution segmentation algorithm to obtain a corresponding image segmentation result; wherein the image segmentation result contains a segmented image of each ground feature in the Sentinel-1 data;

[0041] An extraction module is configured to extract a polarized wave band feature, a texture feature, a water body index feature, and a time SAR feature of the Sentinel-1 data;

[0042] The training module is configured to take the terrain feature, the image segmentation result of the Sentinel-1 data, the polarized wave band feature, the texture feature, the water body index feature, and the time SAR feature as inputs of an initial random forest model corresponding to the Sentinel-1 data, and perform model training on the initial random forest model by using the training sample, so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish the water body region from the non-water body region in the Sentinel-1 data.

[0043] In a third aspect, an electronic device is provided, which includes a processor, a memory, and a bus. The memory stores machine readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory through the bus. The machine readable instructions are executed by the processor to perform the steps in any possible implementation of the first aspect.

[0044] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps in any possible implementation of the first aspect are performed.

[0045] The flood monitoring method, device, electronic device, and computer readable storage medium provided by the embodiments of the present application can improve the accuracy of the random forest model in identifying water body regions and non-water body regions by combining the terrain features of the flood occurrence region, the polarized wave band SAR features of the Sentinel-1 data, the texture features, the water body index features, the time SAR features, and the segmented images of each ground feature in the Sentinel-1 data when training the initial random forest model.

[0046] In order to make the above objectives, characteristics and advantages of the present application more apparent and understandable, the following will describe preferred embodiments in detail, and the accompanying drawings will be described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0048] Figure 1 A flowchart of a flood monitoring method provided by an embodiment of the present application is shown;

[0049] Figure 2 A schematic diagram of an image segmentation result is shown according to an embodiment of the present application;

[0050] Figure 3 A structural schematic diagram of a flood monitoring device is shown according to an embodiment of the present application;

[0051] Figure 4 A structural schematic diagram of an electronic device is shown according to an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the accompanying drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0053] Machine learning classification techniques have been gradually applied to flood monitoring based on SAR images and have achieved high extraction accuracy. Machine learning methods can efficiently process high-dimensional data, make full use of image feature information, and the trained model can be applied to multiple images of the same type. Therefore, some researchers have devoted themselves to the study of machine learning methods and flood monitoring using multi-temporal SAR images. The use of multi-temporal SAR provides a means to better understand the seasonal behavior of different land cover classes. At the same time, time series analysis can be used to obtain multi-temporal features, thereby enabling more detailed extraction of flood-related classes. From the perspective of classification objects, although pixel-based flood information extraction is the most commonly used method, it only considers the information of the pixel itself, ignoring the spatial arrangement, texture, and morphological features of the object, and the fragmented and redundant classification results may not meet the accuracy requirements. The object-oriented method divides the remote sensing image into image objects as the basic unit of classification. After segmentation, the inherent features of each object (such as spectral, texture, and shape features) and the features describing the relationship between objects (such as connectivity and proximity) are used for object classification, and the resulting results have relatively low redundancy and fragmentation, significant improvement in feature classification accuracy, and good improvement in noise problems. Therefore, the object-oriented machine learning method is becoming a very effective image classification method and is being increasingly used in flood monitoring research. Among them, the random forest algorithm has high accuracy and anti-noise interference ability. At the same time, the importance of features can be estimated, and dimensionality reduction can be achieved in high-dimensional feature space. Compared with the current popular similar algorithms, it has faster learning speed and higher accuracy and stability.

[0054] Although many achievements have been made in flood identification using SAR images, it is still challenging to classify water types in flood disaster events from post-disaster remote sensing images due to the fact that the backscattering signal of temporary water bodies is influenced by both the water body and the submerged ground object. From the perspective of classification objects, although pixel-based flood information extraction is the most commonly used method, it only considers the information of the pixel itself, ignoring the spatial arrangement, texture, and morphological features of the object, and the fragmented and redundant classification results may not meet the accuracy requirements.

[0055] In view of the above problems, based on this, the embodiment of the application provides a flood monitoring method, device, electronic equipment and computer readable storage medium, which combines polarization band features, texture features, water body index features, time SAR features and terrain features to train an initial random forest model, thereby improving the accuracy of the random forest model obtained by training in water body area identification. The following embodiments are described.

[0056] In order to facilitate the understanding of the present embodiment, first of all, a flood monitoring method disclosed by the present embodiment is introduced in detail. As Figure 1As shown, the method comprises the following steps S101-S105:

[0057] S101: acquiring digital elevation data of the flood-occurred region, multispectral image data, and time series Sentinel-1 data in a monitoring time period; wherein the monitoring time period is a time period from before the flood disaster to after the flood disaster of the flood-occurred region.

[0058] In this embodiment, the digital elevation data (DEM data) of the flood-occurred region is used to describe the spatial distribution of the terrain elevation of the flood-occurred region.

[0059] The multispectral image data is HJ-2 multispectral verification data, which is a standardized pre-processed data set generated by the HJ-2A / B satellite multispectral load, mainly used to verify the radiation accuracy, geometric positioning accuracy and ground object classification reliability of remote sensing images. The multispectral image data is generated and acquired within a preset time period (for example, within 3 days, including 3 days) after the flood, in other words, the generation time of the multispectral image data of the flood-occurred region is less than the preset time period (for example, 3 days, including 3 days) from the day of the flood. For example, assuming that the flood disaster occurred on July 8, 2021, then the generation and acquisition time of the multispectral image data can be July 10, 2021.

[0060] In this embodiment, the time series Sentinel-1 data of the sample region in the monitoring time period is acquired from the GEE platform. In this embodiment, the acquired time series Sentinel-1 data can be time series Sentinel-1B IW mode GRD product data.

[0061] The time series Sentinel-1 data contains multiple Sentinel-1 data with time sequence. The Sentinel-1 data can be Sentinel-1B IW mode GRD product data. The Sentinel-1B IW mode GRD product data (Sentinel-1B IW mode Level-1 ground range detection (GRD) product data) belongs to Level-1 product in the Sentinel-1 data system, which is a radar intensity image generated by synthesizing aperture radar (SAR) to acquire ground backscatter signals and multi-view processing.

[0062] The Sentinel-1 data contains radar backscatter intensity, geospatial information, and metadata. The radar backscatter intensity is used to reflect the electromagnetic wave reflection characteristics of the ground surface, and can be used to invert soil moisture, water body range, and other ground surface parameters. The geospatial information includes latitude and longitude coordinates, spatial resolution, and coverage range. The metadata records satellite orbit parameters, imaging time, polarization mode (such as VV+VH dual polarization), and processing level information.

[0063] The monitoring time period is a period of time that includes the occurrence of a flood area before and after the occurrence of a flood disaster. For example, assuming that the flood disaster occurred on July 8, 2021, the monitoring time period can be from May 1, 2021 to September 1, 2021 (a total of 124 days). Then, the time series Sentinel-1 data can contain 10 Sentinel-1 data with time series relationship (i.e., with time sequence relationship).

[0064] In this embodiment, the flood occurrence area can be one or multiple. When the flood occurrence area is multiple, each flood occurrence area corresponds to respective digital elevation data, multi-spectral image data, and time series Sentinel-1 data. At this time, the steps in this embodiment are sequentially performed for each flood occurrence area.

[0065] S102: Determine the topographic features of the flood occurrence area according to the digital elevation data; and select a plurality of water sample areas and a plurality of non-water sample areas from the multi-spectral image data as training samples.

[0066] In this embodiment, in order to improve the reliability of random forest classification, the derived topographic features of the digital elevation data are introduced as auxiliary input. The topographic features include the elevation, slope, aspect, and mountain shadow of the flood occurrence area.

[0067] In this embodiment, the Sentinel-1 data is actually a grayscale image, and the water bodies and roads in the image are represented by black color. Therefore, it is difficult to distinguish water bodies in Sentinel-1 data. Based on this, in this embodiment, the multi-spectral image data is used as the training sample. The multi-spectral image data is a multi-spectral image, and the water body area and the non-water body area are represented by different colors. Therefore, the water body area can be labeled based on the multi-spectral image data.

[0068] Specifically, assuming that in this embodiment, the flood occurrence area contains a plurality of water body areas (for example, 500 water body areas) and a plurality of non-water body areas (for example, 1200 non-water body areas). Then, the multi-spectral image data will label 500 water body areas and 1200 non-water body areas.

[0069] A plurality of water body regions (e.g., 50 water body regions) are selected from the multispectral image data as water body sample regions, and a plurality of non-water body regions (e.g., 100 non-water body regions) are selected as non-water body sample regions, and the selected plurality of water body sample regions (e.g., 50 water body sample regions) and the plurality of non-water body sample regions (e.g., 100 non-water body sample regions) are used as training samples.

[0070] The training samples can include the positions of each water body sample region and the water body annotation label corresponding to each water body sample region, and the positions of each non-water body sample region and the non-water body annotation label corresponding to each non-water body sample region.

[0071] The number of selected water body sample regions is less than the total number of water body regions, and similarly, the number of selected non-water body sample regions is less than the total number of non-water body regions.

[0072] In this embodiment, when the plurality of water body sample regions and the plurality of non-water body sample regions are selected from the multispectral image data, the plurality of water body sample regions and the plurality of non-water body sample regions can be randomly selected from the multispectral image data.

[0073] S103: For each Sentinel-1 data in the time series Sentinel-1 data, an image segmentation algorithm is used to perform image segmentation on the Sentinel-1 data to obtain a corresponding image segmentation result; wherein the image segmentation result includes a segmented image of each ground object in the Sentinel-1 data.

[0074] In this embodiment, the ground objects include water bodies, vegetation, and construction land. The image segmentation result includes a segmented image corresponding to each ground object in the flood-affected area, for example, a water body corresponds to a segmented image. The segmented image corresponding to each ground object refers to the area where the ground object is located is segmented from the Sentinel-1 data, and the average value of all pixels in the area where the ground object is located is calculated to obtain the segmented image. As shown in Figure 2

[0075] In this embodiment, each Sentinel-1 data corresponds to a respective image segmentation result.

[0076] In one possible implementation, before step S103 is performed, each Sentinel-1 data can also be preprocessed by the following steps:

[0077] ​For each of the time series Sentinel-1 data, the Sentinel-1 data is preprocessed to obtain preprocessed Sentinel-1 data; wherein the preprocessing includes: applying an orbit file, GRD boundary noise removal, thermal noise removal, radiometric calibration, filtering processing, terrain correction, converting backscatter intensity into decibels;

[0078] The orbit file is used in GRD product data preprocessing to correct the position and speed information of the satellite to eliminate geographical positioning errors caused by satellite orbit parameter deviation. Thermal noise removal is used to eliminate radar system noise (such as thermal noise, sensor inherent noise) and improve the signal-to-noise ratio (default to retain VH and VV polarization channels). Radiometric calibration is used to convert the original digital value (DN) into backscatter coefficient (σ0 or β0) to achieve radiation consistency of different time phase data.

[0079] Terrain correction is performed using SRTM. Specifically, SRTM or higher resolution DEM data is used to correct terrain distortion through Range-Doppler model to generate geocoded images. Backscatter intensity is converted into decibels (dB) through logarithmic scale to enhance the contrast of low reflection areas. A 3x3 pixel window size Refined Lee filter is used to remove speckle noise, suppress the inherent speckle noise of the Sentinel-1 data, and enhance the texture features of the ground objects.

[0080] In one possible implementation, when performing step S103, the Sentinel-1 data is segmented using a multi-resolution segmentation algorithm to obtain the corresponding image segmentation result, which can be performed according to the following steps:

[0081] An optimal segmentation scale estimation tool (ESP2) is called to determine the optimal segmentation scale parameters (such as 80 pixels), the optimal shape factor (such as 0,1), and the compactness parameters (such as 0,5) of the preprocessed Sentinel-1 data.

[0082] The determined optimal segmentation scale parameters, optimal shape factor, and compactness parameters are input into the multi-resolution segmentation algorithm, and the preprocessed Sentinel-1 data is segmented using the multi-resolution segmentation algorithm to obtain the image segmentation result.

[0083] S104: Extracting polarized band features, texture features, water index features, and time SAR features of the Sentinel-1 data.

[0084] In one possible implementation, when performing step S104, the following steps S1041-S1044 can be performed:

[0085] S1041: Based on the VV polarization and the VH polarization of the Sentinel-1 data, a polarization ratio, a polarization difference, a polarization sum, and a polarization product are calculated to obtain the polarization band features of the Sentinel-1 data.

[0086] In this embodiment, each Sentinel-1 data corresponds to a VV (vertical transmission vertical reception) polarization and a VH (vertical transmission horizontal reception) polarization. For each Sentinel-1 data, the ratio of the VV polarization to the VH polarization (i.e., the polarization ratio), the difference between the VV polarization and the VH polarization (i.e., the polarization difference), the sum of the VV polarization and the VH polarization (i.e., the polarization sum), and the product of the VV polarization and the VH polarization (i.e., the polarization product) are calculated. The calculated polarization ratio, polarization difference, polarization sum, and polarization product are taken as the polarization band features of the Sentinel-1 data.

[0087] In this embodiment, each Sentinel-1 data corresponds to respective polarization band features.

[0088] S1042: Based on the Gray-Level Co-occurrence Matrix (GLCM), SAR dual-polarization texture features are extracted for the VV polarization and the VH polarization of the Sentinel-1 data respectively to obtain the texture features of the Sentinel-1 data; wherein the texture features are used to reflect the differences of ground objects in space.

[0089] In this embodiment, based on the Gray-Level Co-occurrence Matrix (GLCM), texture features such as Angular Second Moment, Entropy, Mean, Variance difference, Correlation, Homogeneity, and Dissimilarity are calculated for the VV polarization and the VH polarization of the Sentinel-1 data respectively.

[0090] S1043: Based on the VV polarization and the VH polarization of the Sentinel-1 data, a Sentinel-1 Dual-Polarized Water Index (SDWI) is calculated, and the dual-polarization water body index is taken as the water body index feature of the Sentinel-1 data; the water body index feature is used to highlight water bodies and distinguish water bodies from other ground objects except water bodies.

[0091] In this embodiment, the dual-polarization water body index is calculated by the following operation formula:

[0092] K SDWI = ln(10·VV·VH)-8

[0093] wherein, VV represents the VV polarization, VH represents the VH polarization, K SDWI represents the dual-polarization water body index.

[0094] S1044: calculating the time SAR feature of the Sentinel-1 data according to the backscattering data of the Sentinel-1 data and the time series Sentinel-1 data; wherein, the time SAR feature includes the backscattering anomaly and the Z-score; the backscattering anomaly is used to represent the difference between the backscattering during the flood disaster and the average backscattering during the non-flood period; the Z-score is used to represent the difference between the backscattering during the flood disaster and the average backscattering of the first monitoring time period.

[0095] In this embodiment, the backscattering anomaly is calculated by the following formula:

[0096]

[0097] wherein, Δσ(t) represents the backscattering anomaly; σ(t) is the backscattering data on the day of the flood disaster; represents the average backscattering data during the non-flood period (i.e. excluding the day of the flood disaster).

[0098] The Z-score is calculated by the following formula:

[0099]

[0100] wherein, Z represents the Z-score; std(σ) represents the standard deviation of the backscattering data of all the first Sentinel-1B IW mode GRD product data.

[0101] S105: taking the terrain feature, the image segmentation result of the Sentinel-1 data, the polarized wave band feature, the texture feature, the water body index feature, and the time SAR feature as the input of the initial random forest model corresponding to the Sentinel-1 data, using the training sample to perform model training on the initial random forest model, so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish the water body area and the non-water body area in the Sentinel-1 data.

[0102] In this embodiment, each Sentinel-1 data corresponds to an initial random forest model, and each Sentinel-1 data is used to train the initial random forest model corresponding to the Sentinel-1 data to obtain a random forest model corresponding to the Sentinel-1 data. The trained random forest model can be used to determine each water body region and non-water body region in the Sentinel-1 data.

[0103] In a possible implementation, when the initial random forest model corresponding to each Sentinel-1 data is trained in step S105, the following steps can be performed:

[0104] S1051: input the terrain feature, and the polarization band feature, the texture feature, the water body index feature, and the time SAR feature of the Sentinel-1 data into the initial random forest model corresponding to the Sentinel-1 data, and determine the importance ranking of each feature input by the initial random forest model.

[0105] In this embodiment, the importance of the features (i.e., the polarization band feature, the texture feature, the water body index feature, the time SAR feature, and the terrain feature) directly affects the classification accuracy of the initial random forest model. The importance ranking of each feature is calculated and optimally selected, which can improve the classification speed of the initial random forest model and reduce data redundancy.

[0106] In this embodiment, the importance ranking of each feature corresponding to each Sentinel-1 data can be different.

[0107] S1042: according to the importance ranking of each feature, a feature is sequentially added to the initial random forest model from high to low to determine the classification accuracy of the initial random forest model after adding each feature.

[0108] For example, it is assumed that the importance ranking of each feature corresponding to one Sentinel-1 data from high to low is: the polarization band feature, the texture feature, the water body index feature, the time SAR feature, and the terrain feature.

[0109] At this time, the polarization band feature corresponding to the Sentinel-1 data is fused into the Sentinel-1 data, and the fused Sentinel-1 data and the image segmentation result corresponding to the Sentinel-1 data are input into the initial random forest model corresponding to the Sentinel-1 data. Based on the training sample, the classification accuracy of the initial random forest model at this time is calculated.

[0110] Next, the polarization band features and texture features corresponding to the Sentinel-1 data are fused into the Sentinel-1 data, and the fused Sentinel-1 data and the image segmentation result corresponding thereto are input into the initial random forest model corresponding to the Sentinel-1 data. Based on the training samples, the classification accuracy of the initial random forest model at this time is calculated.

[0111] Similarly, each time one feature is fused, the classification accuracy of the initial random forest model corresponding to the Sentinel-1 data after adding one feature can be obtained.

[0112] S1053: According to the change of the classification accuracy of the initial random forest model, the features added to the initial random forest model when the classification accuracy is the highest are determined, and the features added when the classification accuracy is the highest are determined as the target features corresponding to the Sentinel-1 data.

[0113] In this embodiment, the screening of the multiple features corresponding to the Sentinel-1 data is completed through steps S1051-S1053, and the optimal feature combination corresponding to the Sentinel-1 data is found.

[0114] S1054: The image segmentation result of the Sentinel-1 data and each target feature are taken as the input of the initial random forest model corresponding to the Sentinel-1 data, and the initial random forest model is trained using the training samples.

[0115] In this embodiment, after the optimal feature combination (i.e., at least one target feature) corresponding to the Sentinel-1 data is found, the initial random forest model corresponding to the Sentinel-1 data is trained next. When the initial random forest model corresponding to the Sentinel-1 data is trained, the specific process can be:

[0116] For each Sentinel-1 data, the feature values of the target features corresponding to the Sentinel-1 data are fused in the Sentinel-1 data to obtain fused Sentinel-1 data;

[0117] The fused Sentinel-1 data and the image segmentation result corresponding to the Sentinel-1 data are input into the initial random forest model to be trained, and the initial classification result of the water sample area and the non-water sample area in the Sentinel-1 data is output by the initial random forest model.

[0118] Based on the training samples, a loss value of the training samples and the initial classification result is calculated, and the loss value is used for model training of the initial random forest model; wherein the training samples include the positions of each water body sample area and the water body annotation label corresponding to each water body sample area, and the positions of each non-water body sample area and the non-water body annotation label corresponding to each non-water body sample area.

[0119] When the preset model training stop condition is met, the model training of the initial random model is stopped, and a trained random forest model corresponding to the Sentinel-1 data is obtained. The trained random forest model has the ability to distinguish water body areas and non-water body areas in the Sentinel-1 data.

[0120] In a possible implementation, for each Sentinel-1 data, after obtaining the trained random forest model corresponding to the Sentinel-1 data, each water body area and non-water body area in the Sentinel-1 data can be classified according to the following steps S201-S202:

[0121] S201: The feature values of each of the target features corresponding to the Sentinel-1 data are fused into the Sentinel-1 data, and fused Sentinel-1 data is obtained.

[0122] S202: The fused Sentinel-1 data and the image segmentation result corresponding to the Sentinel-1 data are input into the random forest model, and based on the random forest model, water body areas and non-water body areas in the fused Sentinel-1 data are identified.

[0123] In this embodiment, for each Sentinel-1 data, part of the water body sample areas (for example, 50 water body sample areas) and non-water body sample areas (for example, 100 non-water body sample areas) in the Sentinel-1 data are used to train the initial random forest model corresponding to the Sentinel-1 data, and after the training is completed, the trained random forest model can distinguish all water body areas (for example, 500 water body sample areas) and non-water body areas (for example, 1200 non-water body sample areas) in the Sentinel-1 data.

[0124] In a possible implementation, after step S202 is performed and the identification result corresponding to each Sentinel-1 data is obtained, the following steps S203-S209 can also be performed:

[0125] S203: Based on the identified water body region and non-water body region in the fused Sentinel-1 data, a binary classification image is generated for distinguishing the water body region and the non-water body region.

[0126] In this embodiment, the water body region in the binary classification image is represented by one color, and the non-water body region is represented by another color.

[0127] S204: After obtaining the respective binary classification image corresponding to each Sentinel-1 data, for each pixel in the binary classification image, the frequency of the pixel being identified as a water body is calculated; wherein the frequency reflects the duration of the water body at the pixel in the monitoring period.

[0128] In this embodiment, the pixel refers to the smallest unit of area in the binary classification image. When calculating the frequency of each pixel being identified as a water body in each binary classification image, the following formula can be used:

[0129]

[0130] wherein N W represents the number of times the pixel W is identified as a water body; N a represents the total number of Sentinel-1 data; f W represents the frequency of the pixel W in each binary classification image being identified as a water body.

[0131] In this embodiment, the greater the frequency corresponding to each pixel, the longer the duration of the water body at the pixel; conversely, the smaller the frequency, the shorter the duration of the water body at the pixel.

[0132] S205: Select a plurality of pixels as sample pixels from all pixels, and determine whether each sample pixel is a permanent water body pixel or a flood pixel in the real situation.

[0133] In this embodiment, each pixel corresponds to a location in the flood-occurring region. In order to identify the permanent water body region (i.e., the region with water body before the flood occurs) and the flood region (i.e., the region without water body before the flood occurs and with water body after the flood occurs) in the flood-occurring region, a plurality of pixels (e.g., hundreds of pixels) are selected as sample pixels from all pixels (e.g., tens of thousands of pixels). Since the number of sample pixels is limited, the location corresponding to each sample pixel can be used to determine whether the sample pixel is a permanent water body pixel or a flood pixel in the real situation.

[0134] The sample pixel is a permanent water body pixel in the real situation refers to the position corresponding to the sample pixel in the flood area is a permanent water body area. The sample pixel is a flood pixel in the real situation refers to the position corresponding to the sample pixel in the flood area is a flood area.

[0135] S206: For each preset frequency threshold, compare the respective frequency of each sample pixel with the preset frequency threshold to infer that each sample pixel is a permanent water body pixel or a flood pixel based on the preset frequency threshold.

[0136] In this embodiment, the value range of the preset frequency threshold can be 0.1-1, and if the step is 0.1, then the value of the preset frequency threshold can be 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1 respectively.

[0137] Taking the preset frequency threshold 0.2 as an example, as shown in Table 1 below, assuming that there are 8 sample pixels (a-h), when inferring that each sample pixel is a permanent water body pixel or a flood pixel based on the preset frequency threshold 0.2:

[0138] The frequency 0.1 of the sample pixel a is less than or equal to the preset frequency threshold 0.2, so the sample pixel a is inferred to be a flood pixel;

[0139] The frequency 0.2 of the sample pixel b is less than or equal to the preset frequency threshold 0.2, so the sample pixel b is inferred to be a flood pixel;

[0140] The frequency 0.3 of the sample pixel c is greater than the preset frequency threshold 0.2, so the sample pixel c is inferred to be a permanent water body pixel;

[0141] The frequency 0.4 of the sample pixel d is greater than the preset frequency threshold 0.2, so the sample pixel d is inferred to be a permanent water body pixel;

[0142] The frequency 0.5 of the sample pixel e is greater than the preset frequency threshold 0.2, so the sample pixel e is inferred to be a permanent water body pixel;

[0143] The frequency 0.6 of the sample pixel f is greater than the preset frequency threshold 0.2, so the sample pixel f is inferred to be a permanent water body pixel;

[0144] The frequency 0.2 of the sample pixel g is less than or equal to the preset frequency threshold 0.2, so the sample pixel g is inferred to be a flood pixel;

[0145] The frequency 0.1 of the sample pixel h is less than or equal to the preset frequency threshold 0.2, so the sample pixel h is inferred to be a flood pixel.

[0146]

[0147] Table 1

[0148] S207: According to the presumed each sample pixel being a permanent water body pixel or a flood pixel, and the determined each sample pixel being a permanent water body pixel or a flood pixel in the real situation, the accuracy of the preset frequency threshold for distinguishing the permanent water body pixel and the flood pixel is calculated.

[0149] In combination with the foregoing embodiments, the accuracy of the preset frequency threshold for distinguishing the permanent water body pixel and the flood pixel can be calculated as 7 / 8=85.7%.

[0150] S208: According to the accuracy of each preset frequency threshold, the preset frequency threshold with the highest accuracy is screened out as a target preset frequency threshold.

[0151] In this embodiment, after the accuracy of each preset frequency threshold is calculated, the preset frequency threshold with the highest accuracy is screened out from the preset frequency thresholds as the target preset frequency threshold. For example, the target preset frequency threshold can be 0.3.

[0152] S209: For each pixel of the binary classification image, the frequency of the pixel is compared with the target preset frequency threshold. If the frequency of the pixel is greater than the target preset frequency threshold, it indicates that the pixel is a permanent water body pixel. If the frequency of the pixel is less than or equal to the target preset frequency threshold, it indicates that the pixel is a flood pixel.

[0153] In this embodiment, after the target preset frequency threshold is screened out, each pixel in the binary classification image can be classified according to the target preset frequency threshold to determine whether each pixel is a permanent water body pixel or a flood pixel, that is, the position corresponding to each pixel in the flood area is classified to determine whether each position is a permanent water body area or a flood area.

[0154] Specifically, when the frequency of the pixel is greater than the target preset frequency threshold, it indicates that the pixel is a permanent water body pixel, and also indicates that the position corresponding to the pixel in the flood area is a permanent water body area. When the frequency of the pixel is less than or equal to the target preset frequency threshold, it indicates that the pixel is a flood pixel, and also indicates that the position corresponding to the pixel in the flood area is a flood area.

[0155] Based on the same technical concept, the embodiments of the present application also provide a flood monitoring device, as shown in the following table. Figure 3 The device comprises:

[0156] The acquisition module 301 is configured to acquire digital elevation data of a flood occurrence area, multispectral image data, and time series Sentinel-1 data in a monitoring time period; wherein the monitoring time period is a time period from before the flood occurrence area occurs a flood disaster to after the flood occurrence area occurs a flood disaster.

[0157] The determination module 302 is configured to determine a topographic feature of the flood occurrence area according to the digital elevation data, and select a plurality of water body sample areas and a plurality of non-water body sample areas from the multispectral image data as training samples.

[0158] The segmentation module 303 is configured to, for each of the time series Sentinel-1 data, perform image segmentation on the Sentinel-1 data by using a multi-resolution segmentation algorithm to obtain a corresponding image segmentation result; wherein the image segmentation result includes a segmented image of each ground object in the Sentinel-1 data.

[0159] The extraction module 304 is configured to extract a polarized band feature, a texture feature, a water body index feature, and a time SAR feature of the Sentinel-1 data.

[0160] The training module 305 is configured to use the topographic feature, the image segmentation result of the Sentinel-1 data, the polarized band feature, the texture feature, the water body index feature, and the time SAR feature as an input of an initial random forest model corresponding to the Sentinel-1 data, and perform model training on the initial random forest model by using the training samples, so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish water body areas and non-water body areas in the Sentinel-1 data.

[0161] Optionally, the device further comprises:

[0162] The preprocessing module is configured to, before the segmentation module 303 performs image segmentation on the Sentinel-1 data by using a multi-resolution segmentation algorithm to obtain a corresponding image segmentation result, pre-process the Sentinel-1 data to obtain pre-processed Sentinel-1 data; wherein the preprocessing includes: applying an orbit file, removing GRD boundary noise, removing thermal noise, radiometric calibration, filtering processing, terrain correction, and converting backscattering intensity into decibels.

[0163] When the segmentation module 303 is configured to perform image segmentation on the Sentinel-1 data by using a multi-resolution segmentation algorithm to obtain a corresponding image segmentation result, the segmentation module 303 is specifically configured to:

[0164] The optimal segmentation scale estimation tool is called to determine the optimal segmentation scale parameter, the shape factor optimal value and the compactness parameter of the preprocessed Sentinel-1 data;

[0165] The determined optimal segmentation scale parameter, the shape factor optimal value and the compactness parameter are input into a multi-resolution segmentation algorithm, and the preprocessed Sentinel-1 data is segmented by using the multi-resolution segmentation algorithm to obtain an image segmentation result.

[0166] Optionally, the extraction module 304 is specifically configured to:

[0167] Based on the VV polarization and the VH polarization of the Sentinel-1 data, a polarization ratio, a polarization difference, a polarization sum and a polarization product are calculated to obtain the polarization band feature of the Sentinel-1 data.

[0168] Based on the gray level co-occurrence matrix, SAR dual-polarization texture features of the VV polarization and the VH polarization of the Sentinel-1 data are extracted respectively to obtain the texture feature of the Sentinel-1 data; wherein the texture feature is used to reflect the difference of ground objects in space.

[0169] Based on the VV polarization and the VH polarization of the Sentinel-1 data, a dual-polarization water body index is calculated, and the dual-polarization water body index is used as the water body index feature of the Sentinel-1 data; the water body index feature is used to highlight the water body and distinguish the water body from other ground objects.

[0170] According to the backscattering data of the Sentinel-1 data and the time series Sentinel-1 data, a time SAR feature of the Sentinel-1 data is calculated; wherein the time SAR feature includes a backscattering anomaly and a Z-score; the backscattering anomaly is used to represent the difference between the backscattering during the flood disaster and the average backscattering during the non-flood period; the Z-score is used to represent the difference between the backscattering during the flood disaster and the average backscattering of the monitoring period.

[0171] Optionally, the terrain features include elevation, slope, slope direction and mountain shadow.

[0172] Optionally, the training module 305 is configured to input the terrain feature, the image segmentation result of the Sentinel-1 data, the polarized wave band feature, the texture feature, the water body index feature, and the time SAR feature of the Sentinel-1 data into an initial random forest model corresponding to the Sentinel-1 data, and perform model training on the initial random forest model by using the training sample, so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish the water body region and the non-water body region in the Sentinel-1 data, and specifically configured to:

[0173] input the terrain feature, the polarized wave band feature, the texture feature, the water body index feature, and the time SAR feature of the Sentinel-1 data into the initial random forest model corresponding to the Sentinel-1 data, and determine the importance ranking of each feature input by the initial random forest model;

[0174] add a feature in the initial random forest model in turn from high to low according to the importance ranking of each feature, so as to determine the classification accuracy of the initial random forest model after adding each feature;

[0175] determine the features added in the initial random forest model when the classification accuracy is the highest according to the change of the classification accuracy of the initial random forest model, and determine the features added when the classification accuracy is the highest as the target features corresponding to the Sentinel-1 data;

[0176] input the image segmentation result of the Sentinel-1 data and each target feature as the input of the initial random forest model corresponding to the Sentinel-1 data, and perform model training on the initial random forest model by using the training sample.

[0177] Optionally, the device further comprises:

[0178] a fusion module configured to, for each Sentinel-1 data, after obtaining the trained random forest model corresponding to the Sentinel-1 data, fuse the feature values of each target feature corresponding to the Sentinel-1 data into the Sentinel-1 data to obtain fused Sentinel-1 data;

[0179] an identification module configured to input the fused Sentinel-1 data and the image segmentation result corresponding to the Sentinel-1 data into the random forest model, and identify the water body region and the non-water body region in the fused Sentinel-1 data based on the random forest model.

[0180] Optionally, the apparatus further comprises:

[0181] a generating module configured to generate, based on the identified water body region and non-water body region in the fused Sentinel-1 data, a binary image for distinguishing the water body region and the non-water body region;

[0182] a first calculating module configured to, after obtaining the respective binary image corresponding to each Sentinel-1 data, calculate, for each pixel in the binary image, a frequency of being identified as a water body; wherein the frequency reflects a duration of the water body at the pixel in the monitoring time period;

[0183] a selecting module configured to select a plurality of pixels from all the pixels as sample pixels, and determine whether each of the sample pixels is a permanent water body pixel or a flood pixel in a real situation;

[0184] a first comparing module configured to, for each preset frequency threshold, compare the respective frequency of each of the sample pixels with the preset frequency threshold, to infer whether each of the sample pixels is a permanent water body pixel or a flood pixel based on the preset frequency threshold;

[0185] a second calculating module configured to calculate, according to the inferred permanent water body pixel or flood pixel of each of the sample pixels and the determined permanent water body pixel or flood pixel of each of the sample pixels in the real situation, an accuracy of the preset frequency threshold for distinguishing the permanent water body pixel and the flood pixel;

[0186] a screening module configured to screen, according to the respective accuracy of each preset frequency threshold, a preset frequency threshold with the highest accuracy as a target preset frequency threshold;

[0187] a second comparing module configured to, for each of the pixels in the binary image, compare the frequency of the pixel with the target preset frequency threshold, and if the frequency of the pixel is greater than the target preset frequency threshold, it indicates that the pixel is a permanent water body pixel; and if the frequency of the pixel is less than or equal to the target preset frequency threshold, it indicates that the pixel is a flood pixel.

[0188] Figure 4 A structural schematic diagram of an electronic device provided by the embodiment of the present application, comprising: a processor 401, a memory 402 and a bus 403, the memory 402 stores machine readable instructions executable by the processor 401, when the electronic device runs the above information processing method, the processor 401 and the memory 402 communicate through the bus 403, the processor 401 executes the machine readable instructions to execute the method steps in the embodiment one.

[0189] The embodiment of the present application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is run by a processor to execute the method steps in the embodiment one.

[0190] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described apparatus, electronic device and computer readable storage medium can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0191] In several embodiments provided in the present application, it should be understood that the disclosed method, apparatus, electronic device and computer readable storage medium can be implemented by other manners. The apparatus embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed units can be indirect coupling or communication connection through some communication interfaces, apparatuses or modules, and can be electrical, mechanical or other forms.

[0192] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0193] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0194] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present application essentially or the parts of the prior art that make contributions or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0195] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present application, which are used to illustrate the technical solutions of the present application, but not to limit them. The protection scope of the present application is not limited thereto. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present application, or make equivalent replacements to some of the technical features. The modifications, changes or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A flood monitoring method, characterized in that, include: Digital elevation data, multispectral image data, and time-series Sentinel-1 data for the flooded area are acquired; wherein, the monitoring period is the time from before the flood disaster to after the flood disaster in the flooded area; Based on the digital elevation data, the topographic features of the flooded area are determined; and multiple water body sample areas and multiple non-water body sample areas are selected from the multispectral image data as training samples. For each Sentinel-1 data frame in the time series Sentinel-1 data, a multi-resolution segmentation algorithm is used to segment the Sentinel-1 data frame to obtain the corresponding image segmentation result; wherein, the image segmentation result contains the segmented image of each ground feature in the Sentinel-1 data. The polarization band features, texture features, water index features, and temporal SAR features of this Sentinel-1 data were extracted. The terrain features, image segmentation results, polarization band features, texture features, water index features, and temporal SAR features of the Sentinel-1 data are used as inputs to the initial random forest model corresponding to the Sentinel-1 data. The initial random forest model is trained using the training samples so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish between water areas and non-water areas in the Sentinel-1 data.

2. The method according to claim 1, characterized in that, Before using a multi-resolution segmentation algorithm to segment the Sentinel-1 data and obtain the corresponding image segmentation result, the process also includes: The Sentinel-1 data was preprocessed to obtain preprocessed Sentinel-1 data; wherein, the preprocessing included: applying orbit file, GRD boundary noise removal, thermal noise removal, radiometric calibration, filtering, terrain correction, and converting backscattering intensity into decibels; The method of using a multi-resolution segmentation algorithm to segment the Sentinel-1 data to obtain the corresponding image segmentation results includes: The optimal segmentation scale estimation tool is invoked to determine the optimal segmentation scale parameters, optimal shape factor values, and compactness parameters for the preprocessed Sentinel data. The determined optimal segmentation scale parameter, the optimal value of the shape factor, and the compactness parameter are input into the multi-resolution segmentation algorithm. The multi-resolution segmentation algorithm is then used to perform image segmentation on the preprocessed Sentinel-1 data to obtain the image segmentation result.

3. The method according to claim 1, characterized in that, The extraction of polarization band features, texture features, water index features, and temporal SAR features from this Sentinel-1 data includes: Based on the VV polarization and VH polarization of this Sentinel-1 data, the polarization ratio, polarization difference, polarization sum, and polarization product are calculated to obtain the polarization band characteristics of this Sentinel-1 data. Based on the gray-level co-occurrence matrix, SAR dual-polarization texture features were extracted from the VV polarization and VH polarization of the Sentinel-1 data to obtain the texture features of the Sentinel-1 data; wherein, the texture features are used to reflect the spatial differences of ground features; Based on the VV polarization and VH polarization of this Sentinel-1 data, a dual-polarization water body index is calculated, and this dual-polarization water body index is used as the water body index feature of this Sentinel-1 data; the water body index feature is used to highlight water bodies to distinguish them from other land features. Based on the Sentinel-1 data and the backscatter data of the time-series Sentinel-1 data, the temporal SAR features of the Sentinel-1 data are calculated; wherein, the temporal SAR features include backscatter anomalies and Z-scores; the backscatter anomalies are used to characterize the difference between backscatter during flood disasters and the average backscatter during non-flood periods; the Z-scores are used to characterize the difference between backscatter during flood disasters and the average backscatter during the monitoring period.

4. The method according to claim 1, characterized in that, The terrain features include elevation, slope, aspect, and mountain shadow.

5. The method according to claim 1, characterized in that, The process of using the terrain features, image segmentation results of the Sentinel-1 data, polarization band features, texture features, water index features, and temporal SAR features as inputs to the initial random forest model corresponding to the Sentinel-1 data, and training the initial random forest model using the training samples, includes: The terrain features, as well as the polarization band features, texture features, water index features, and time SAR features of the Sentinel-1 data, are input into the initial random forest model corresponding to the Sentinel-1 data. The importance ranking of each input feature is determined through the initial random forest model. Based on the importance of each feature, one feature is added to the initial random forest model in descending order of importance to determine the classification accuracy of the initial random forest model after each feature is added. Based on the changes in the classification accuracy of the initial random forest model, the features added to the initial random forest model when the classification accuracy is highest are determined, and the features added when the classification accuracy is highest are determined as the target features corresponding to the Sentinel-1 data. The image segmentation result corresponding to the Sentinel-1 data and each of the target features are used as inputs to the initial random forest model corresponding to the Sentinel-1 data. The initial random forest model is trained using the training samples.

6. The method according to claim 5, characterized in that, The method further includes: For each Sentinel-1 dataset, after obtaining the trained random forest model corresponding to that Sentinel-1 dataset, the feature values ​​of each target feature corresponding to that Sentinel-1 dataset are fused into that Sentinel-1 dataset to obtain the fused Sentinel-1 dataset. The fused Sentinel-1 data and the corresponding image segmentation result are input into the random forest model. Based on the random forest model, water and non-water regions in the fused Sentinel-1 data are identified.

7. The method according to claim 6, characterized in that, The method further includes: Based on the identified water and non-water regions in the fused Sentinel-1 data, a binary classification image is generated to distinguish between the water and non-water regions; After obtaining the binary classification image corresponding to each Sentinel-1 data, for each pixel in the binary classification image, the frequency at which the pixel is identified as a water body is calculated; wherein, the frequency is used to reflect the duration of the water body at that pixel during the monitoring period. Multiple pixels are selected from all the pixels as sample pixels, and each sample pixel is determined to be a permanent water body pixel or a flood pixel in reality. For each preset frequency threshold, the frequency corresponding to each sample pixel is compared with the preset frequency threshold, so as to infer whether each sample pixel is a permanent water body pixel or a flood pixel based on the preset frequency threshold. Based on the inferred fact that each of the sample pixels is a permanent water body pixel or a flood pixel, and the determined fact that each of the sample pixels is a permanent water body pixel or a flood pixel, the accuracy of the preset frequency threshold used to distinguish between permanent water body pixels and flood pixels is calculated. Based on the accuracy of each preset frequency threshold, the preset frequency threshold with the highest accuracy is selected as the target preset frequency threshold. For each pixel in the binary classification image, the frequency of the pixel is compared with the target preset frequency threshold. If the frequency of the pixel is greater than the target preset frequency threshold, the pixel is considered a permanent water body pixel. If the frequency of the pixel is less than or equal to the target preset frequency threshold, the pixel is considered a flood pixel.

8. A flood monitoring device, characterized in that, include: The acquisition module is used to acquire digital elevation data, multispectral image data, and time-series Sentinel-1 data of the flooded area within the monitoring period; wherein, the monitoring period is the time period from before the flood disaster to after the flood disaster in the flooded area; The determination module is used to determine the topographic features of the flooded area based on the digital elevation data; and to select multiple water body sample areas and multiple non-water body sample areas from the multispectral image data as training samples. The segmentation module is used to segment each Sentinel-1 data frame in the time series Sentinel-1 data using a multi-resolution segmentation algorithm to obtain the corresponding image segmentation result; wherein, the image segmentation result includes the segmented image of each ground feature in the Sentinel-1 data. The extraction module is used to extract the polarization band features, texture features, water index features, and temporal SAR features of this Sentinel-1 data. The training module is used to take the terrain features, the image segmentation results of the Sentinel-1 data, the polarization band features, the texture features, the water index features, and the temporal SAR features as inputs to the initial random forest model corresponding to the Sentinel-1 data. Using the training samples, the initial random forest model is trained so that the trained random forest model corresponding to the Sentinel-1 data has the ability to distinguish between water areas and non-water areas in the Sentinel-1 data.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the memory via the bus, and the machine-readable instructions, when executed by the processor, perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method as described in any one of claims 1 to 7.

Images