A method, device, equipment and storage medium for analyzing flood damage in cultivated land

Abstract

The invention discloses a method, device, equipment and storage medium for analyzing farmland flood disasters. The method comprises the following steps: obtaining multi-source remote sensing data of farmland flood disaster areas; inputting multispectral remote sensing images and DEM data into a farmland element extraction network to obtain farmland element spatial distribution results; extracting pre-disaster radar images and post-disaster radar images based on a convolutional neural network to obtain newly added water body areas; superimposing farmland element spatial distribution results and newly added water body areas to obtain farmland fully submerged areas; dividing river data in DEM data based on a Thiessen polygon method to obtain sub-catchment areas; superimposing farmland fully submerged areas and sub-catchment areas to calculate the average elevation of farmland fully submerged areas in each sub-catchment area; and performing elevation comparison on each sub-catchment area to determine farmland below the average elevation as farmland semi-submerged areas. The method can extract semi-submerged areas and improve the accuracy of flood disaster monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of disaster monitoring, and in particular to a method, device, equipment and storage medium for analyzing flood damage to cultivated land. Background Art

[0002] At present, flood disasters are one of the main natural disasters in agriculture, with strong destructive power and huge economic losses. In order to ensure rapid emergency response after agricultural disasters, it is necessary to promptly and quickly count the location, region, area and other information of the disaster-stricken areas. Compared with the field sampling survey method, remote sensing technology has the characteristics of low cost, large area and high efficiency in flood disaster emergency monitoring, and has become one of the main technical means of flood disaster emergency monitoring.

[0003] In the existing technology, farmland flood disaster monitoring mainly includes farmland extraction, new water body extraction, and farmland flood disaster analysis. Among them, farmland extraction mainly uses multispectral remote sensing images to extract the spectral and spatial texture characteristics of farmland, and uses classifiers such as random forest and support vector machine for classification; new water body extraction mainly uses synthetic aperture radar (SAR) images, taking advantage of the relatively low backscattering coefficient of water bodies in SAR images, using threshold segmentation or supervised classification methods to extract water bodies, and using digital elevation models (DEM) to remove the influence of mountain shadows, and taking the water body extracted from SAR images at the time of the disaster minus the water body extracted from SAR images before the disaster as the new water body; farmland flood disaster analysis superimposes the new water body on the extracted farmland, and takes the overlapping area of ​​the two as the farmland disaster area.

[0004] However, when the crops planted on the cultivated land are in the late growth stage and the planting is relatively dense, there is a certain degree of water accumulation in the cultivated land (semi-flooded area of ​​crops), which is still mainly manifested as vegetation characteristics in the above SAR images. The existing monitoring method directly superimposes the newly added water bodies extracted from SAR images on the cultivated land, and cannot extract the semi-flooded area of ​​crops, resulting in the extracted disaster area being smaller than the actual disaster area, thus affecting the accuracy of flood disaster monitoring. Summary of the invention

[0005] The present invention provides a method, device, equipment and storage medium for analyzing flood damage in cultivated land, so as to solve the problem that the prior art cannot extract the semi-flooded area of ​​crops, resulting in the extracted disaster-stricken area being smaller than the actual disaster-stricken area, thus affecting the accuracy of flood disaster monitoring.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for analyzing flood damage to cultivated land, comprising:

[0007] Acquire multi-source remote sensing data of the flood-affected areas of cultivated land; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images, and post-disaster radar images;

[0008] Inputting the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements;

[0009] Based on a pre-trained convolutional neural network, the pre-disaster radar image and the post-disaster radar image are respectively extracted to obtain a newly added water body area;

[0010] The spatial distribution result of the cultivated land elements and the newly added water body area are superimposed to obtain the cultivated land fully submerged area;

[0011] Dividing the river data in the DEM data based on the Thiessen polygon method to obtain sub-catchment areas;

[0012] Superimposing the fully submerged cultivated land area and the sub-catchment area, and calculating the average elevation of the fully submerged cultivated land area in each sub-catchment area;

[0013] The elevation of each sub-catchment area is compared respectively, and the cultivated land below the average elevation is determined as the semi-flooded area of ​​cultivated land.

[0014] Preferably, the dividing of the river data in the DEM data based on the Thiessen polygon method to obtain sub-catchment areas includes:

[0015] According to the river flow direction grid data, all connected pixel groups belonging to the same drainage basin are obtained, and the pouring points and depression points at the edge of the window are located;

[0016] Identify the catchment area on each of the pour points and construct an initial subcatchment area;

[0017] Extract the DEM data to obtain a water flow direction layer under unfilled conditions, and determine the outlets of each of the initial sub-catchment areas; wherein the point with the lowest ground elevation or the point to which the opposite water flow direction points among all the inclined points with a water flow direction grid change amplitude less than or equal to 45° is the outlet of each sub-catchment area;

[0018] According to the outlets of the initial sub-catchment area, all adjacent outlets are connected to form a triangle, and perpendicular bisectors are drawn for each side of the triangle, and a polygon surrounded by several perpendicular bisectors around each outlet is determined as a sub-catchment area.

[0019] Preferably, the pre-trained convolutional neural network extracts the pre-disaster radar image and the post-disaster radar image respectively to obtain the newly added water body area, including:

[0020] Based on a pre-trained convolutional neural network, the pre-disaster radar image and the post-disaster radar image are respectively extracted to obtain an initial water body area;

[0021] Acquire the satellite's position information when it passes by, and simulate the mountain shadow area when the satellite passes by according to the position information and the DEM data;

[0022] The initial water body area is screened out according to the mountain shadow area to obtain a newly added water body area.

[0023] Preferably, the training process of the farmland feature extraction network includes:

[0024] Based on the multispectral remote sensing image, a deep spatial-spectral fusion cultivated land element extraction network is constructed;

[0025] Constructing a labeled data set of the cultivated land remote sensing image based on the multispectral remote sensing image, the DEM data and a preset cultivated land labeled vector map;

[0026] The farmland feature extraction network is trained according to the labeled data set.

[0027] Preferably, the encoder of the farmland feature extraction network includes a global context aggregation module, and the global context aggregation module includes global context modeling, channel dependency conversion and feature fusion.

[0028] Preferably, the training process of the convolutional neural network includes:

[0029] Annotate water bodies based on the pre-disaster radar images to construct a water body annotation dataset for radar remote sensing images;

[0030] Convolutional neural network is constructed by taking the image space block composed of the marked pixel as the center and the surrounding neighborhood pixels as input;

[0031] The convolutional neural network is trained according to the water body annotation dataset.

[0032] Preferably, the calculation formula of the hill shadow area includes:

[0033] Hs=255×((cos(Z)×cos(S))+(sin(Z)×sin(S)×cos(Az_math-As)))

[0034]

[0035] Where Z is the zenith angle when the satellite passes by, that is, the angle between the direction of the radar satellite transmitting signal and the local ground plane; S is the slope and aspect data of a certain point on the ground calculated from high-precision DEM data; when Hs is 0, it indicates the mountain shadow area that cannot be covered by the signal when the radar satellite passes by; Az_math is the mathematical azimuth converted from the geographical azimuth Az of the satellite orbit.

[0036] In a second aspect, the present invention provides a device for analyzing flood damage to cultivated land, comprising:

[0037] A data acquisition module is used to acquire multi-source remote sensing data of the flood-affected areas of cultivated land; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images and post-disaster radar images;

[0038] An element extraction module is used to input the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements;

[0039] A water body extraction module, used to extract the pre-disaster radar image and the post-disaster radar image respectively based on a pre-trained convolutional neural network to obtain a newly added water body area;

[0040] A total flooding analysis module is used to superimpose the spatial distribution results of the cultivated land elements and the newly added water body area to obtain the cultivated land total flooding area;

[0041] A river channel division module is used to divide the river channel data in the DEM data based on the Thiessen polygon method to obtain a sub-catchment area;

[0042] An elevation calculation module, used for superimposing the fully flooded cultivated land area and the sub-catchment area, and calculating the average elevation of the fully flooded cultivated land area in each sub-catchment area;

[0043] The semi-flooded analysis module is used to compare the elevations of each of the sub-catchment areas and determine the cultivated land below the average elevation as the semi-flooded cultivated land area.

[0044] In a third aspect, the present invention further provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for analyzing farmland flood damage described in any one of the above is implemented.

[0045] In a fourth aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned methods for analyzing farmland flood damage.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention provides a method for analyzing farmland flood disasters, comprising obtaining multi-source remote sensing data of farmland flood disaster areas; wherein the multi-source remote sensing data comprises multispectral remote sensing images, DEM data, pre-disaster radar images and post-disaster radar images; inputting the multispectral remote sensing images and the DEM data into a pre-trained farmland element extraction network to obtain farmland element spatial distribution results; extracting the pre-disaster radar images and the post-disaster radar images based on a pre-trained convolutional neural network respectively to obtain newly added water body areas; superimposing the farmland element spatial distribution results and the newly added water body areas to obtain farmland fully submerged areas; dividing river channel data in the DEM data based on a Thiessen polygon method to obtain sub-watershed areas; superimposing the farmland fully submerged areas and the sub-watershed areas, and calculating the average elevation of the farmland fully submerged areas in each of the sub-watershed areas; performing elevation comparison on each of the sub-watershed areas respectively, and determining the farmland below the average elevation as a farmland semi-submerged area.

[0048] The present invention aims to solve the problem that the current method of directly superimposing the newly added water body and the cultivated land area cannot extract the semi-flooded area of ​​crops. The present invention divides the sub-catchment area in the region through high-precision DEM data, calculates the average elevation of the newly added water body covering the cultivated land area in each sub-catchment area, and sets the cultivated land below the average elevation in the sub-catchment area as the disaster-affected cultivated land area. The present invention can realize the extraction of the semi-flooded cultivated land area, accurately output the disaster-affected cultivated land area, and improve the accuracy of flood disaster monitoring. At the same time, the present invention simulates the mountain shadow according to the incident angle and altitude angle when the satellite passes through the high-precision DEM data, and uses post-processing to eliminate the mountain shadow area mistakenly classified as a water body, thereby further improving the accuracy of the water body extraction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the flow of a method for analyzing farmland flood damage provided by the first embodiment of the present invention;

[0050] Figure 2 It is a schematic diagram of the structure of the global context aggregation module;

[0051] Figure 3 It is a schematic diagram of the pixels in the neighborhood of the water body selected for marking;

[0052] Figure 4 It is a schematic diagram of the structure of a device for analyzing farmland flood damage provided by the second embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Reference Figure 1 The first embodiment of the present invention provides a method for analyzing flood damage to cultivated land, comprising the following steps:

[0055] S11, obtaining multi-source remote sensing data of the farmland flood-affected area; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images, and post-disaster radar images;

[0056] S12, inputting the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements;

[0057] S13, extracting the pre-disaster radar image and the post-disaster radar image respectively based on a pre-trained convolutional neural network to obtain a newly added water body area;

[0058] S14, superimposing the spatial distribution result of the cultivated land elements and the newly added water body area to obtain the cultivated land fully submerged area;

[0059] S15, dividing the river data in the DEM data based on the Thiessen polygon method to obtain sub-catchment areas;

[0060] S16, superimposing the fully submerged cultivated land area and the sub-catchment area, and calculating the average elevation of the fully submerged cultivated land area in each sub-catchment area;

[0061] S17, performing elevation comparison on each of the sub-catchment areas, and determining the cultivated land below the average elevation as a semi-flooded cultivated land area.

[0062] In step S11, firstly, multi-source remote sensing data of the flood-affected area of ​​cultivated land is obtained; wherein, the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images and post-disaster radar images. Exemplarily, the multi-spectral remote sensing images include cloud-free images fused with long-time series multi-spectral remote sensing images, such as Gaofen-1 and 6, and Sentinel-2. The pre-disaster radar images and post-disaster radar images are satellite radar remote sensing images, such as Gaofen-3, Sentinel-1 and other satellite data. The DEM data is high-precision DEM data.

[0063] In step S12, the multispectral remote sensing image and the DEM data are input into a pre-trained farmland element extraction network to obtain a farmland element spatial distribution result. The training process of the farmland element extraction network includes:

[0064] Based on the multispectral remote sensing image, a deep spatial-spectral fusion cultivated land element extraction network is constructed;

[0065] Constructing a labeled data set of the cultivated land remote sensing image based on the multispectral remote sensing image, the DEM data and a preset cultivated land labeled vector map;

[0066] The farmland feature extraction network is trained according to the labeled data set.

[0067] In this embodiment, based on cloud-free images fused from long-term series multispectral remote sensing images, such as Gaofen-1 and 6, and Sentinel-2, a farmland feature extraction network with deep spatial-spectral information fusion is constructed to accurately extract the spatial distribution of farmland features.

[0068] Specifically, first, based on the cloud-free images and high-precision DEM data fused from long-time series multispectral remote sensing images, combined with the cultivated land annotation vector map in the land change survey, a cultivated land remote sensing image annotation dataset is constructed. Among them, the present invention combines multispectral remote sensing images and DEM data to alleviate the misclassification of cultivated land and forest land, and the DEM slope can be used as effective information to distinguish between flat cultivated land and steep forests in mountainous areas. Then, based on the constructed cultivated land remote sensing image annotation dataset, the constructed deep spatial-spectral fusion classification full convolutional network, such as ResUNet, PSPNet and other network models, is trained. Among them, in order to solve the problem of large spectral variability within the land feature class, the present invention embeds a global context aggregation module in the encoder of the deep spatial-spectral fusion classification network, and alleviates the variance within the cultivated land class by introducing context information. The encoder of the cultivated land feature extraction network includes a global context aggregation module, and the global context aggregation module includes global context modeling, channel dependency conversion and feature fusion, and its network structure is as follows Figure 2 As shown in Figure 2, this module assumes that global context information is less affected by spatial position and converts global context information into dependencies between channels. It is divided into three parts: global context modeling, channel dependency conversion, and feature fusion. For example, for the input feature map X∈R C×H×W , the global context feature map of its output Any pixel x in i ∈R C×1×1 , the formula can be expressed as:

[0069]

[0070] In the above formula, x={x i |i=1,…,Np} represents the input data, where N p Indicates the number of pixels in the feature map. represents global context modeling, where α j represents the weight of global attention pooling, and its formula can be expressed as:

[0071]

[0072] In the above formula, W k Represents a 1×1 convolution. represents the channel dependency conversion function, and its formula can be expressed as:

[0073]

[0074] In the above formula, W v1 represents r 1×1 convolutions, LN represents layer normalization, σ ReLU represents the ReLU activation function, W v2 Represents C 1×1 convolutions.

[0075] After the training of the farmland feature extraction network is completed, the multispectral optical remote sensing images and DEM data of the disaster area are input to infer the farmland feature extraction results. The training data and inference data of the network should have the same spatial resolution and band.

[0076] It can be understood that in terms of cultivated land element extraction, the present invention aims at the problems of large intra-class variance of cultivated land spectra and low precision of current classification methods. A fully convolutional deep learning classification network is designed to mine high-level semantic features, deeply integrate the spectral and spatial attribute information of cultivated land elements, and introduce high-precision DEM information to assist cultivated land extraction and classification. This can solve the problem of serious misclassification between cultivated land and woodland and grassland in the classification results of the prior art.

[0077] In step S13, the pre-disaster radar image and the post-disaster radar image are extracted based on the pre-trained convolutional neural network to obtain the newly added water body area, including:

[0078] Based on a pre-trained convolutional neural network, the pre-disaster radar image and the post-disaster radar image are respectively extracted to obtain an initial water body area;

[0079] Acquire the satellite's position information when it passes by, and simulate the mountain shadow area when the satellite passes by according to the position information and the DEM data;

[0080] The initial water body area is screened out according to the mountain shadow area to obtain a newly added water body area.

[0081] In one embodiment, the training process of the convolutional neural network includes:

[0082] Annotate water bodies based on the pre-disaster radar images to construct a water body annotation dataset for radar remote sensing images;

[0083] Convolutional neural network is constructed by taking the image space block composed of the marked pixel as the center and the surrounding neighborhood pixels as input;

[0084] The convolutional neural network is trained according to the water body annotation dataset.

[0085] It should be noted that the present invention is based on radar remote sensing images before and after the flood disaster, uses convolutional neural networks to extract the water body distribution before and after the disaster, and introduces high-precision DEM data to eliminate the influence of mountain shadows.

[0086] In the specific implementation, we first construct a radar remote sensing image water body annotation dataset based on satellite radar remote sensing images, such as Gaofen-3, Sentinel-1 and other satellite data, combined with water system vector data and manual assistance to annotate water bodies. Then, based on the constructed radar remote sensing image water body annotation dataset, we train convolutional neural classification networks, such as AlexNet, ResNet, VGG and other models. The input of the convolutional neural network is an image space block composed of the marked pixel as the center and the surrounding neighborhood pixels. The corresponding label category is the category label of the center pixel. This method can utilize spatial information while improving the ability to retain details. The schematic diagram is shown below. Figure 3 shown.

[0087] After the convolutional neural network training is completed, the pre-disaster radar remote sensing images and post-disaster radar remote sensing images of the flood disaster area are input to calculate the pre-disaster and post-disaster water body distribution maps, and the areas covered by water bodies after the disaster but not covered by water bodies before the disaster are defined as the initial new water body areas. Among them, the training data and inference data of the network should have the same band and the spatial resolution should not be much different.

[0088] Next, the initial newly added water area is screened out according to the mountain shadow area to obtain the newly added water area, that is, the initial newly added water area is subtracted from the mountain shadow area to obtain the final newly added water area of ​​the flood disaster without the mountain shadow. Specifically, the azimuth and zenith angle of the satellite when passing through are obtained by querying the header information of the satellite image, and the shadow coverage area when the satellite passes through is simulated by high-precision DEM data. The formula can be expressed as:

[0089] Hs=255×((cos(Z)×cos(S))+(sin(Z)×sin(S)×cos(Az_math-As)))

[0090]

[0091] Where Z is the zenith angle when the satellite passes by, that is, the angle between the direction of the radar satellite transmitting signal and the local ground plane; S is the slope and aspect data of a certain point on the ground calculated from high-precision DEM data; when Hs is 0, it indicates the mountain shadow area that cannot be covered by the signal when the radar satellite passes by; Az_math is the mathematical azimuth converted from the geographical azimuth Az of the satellite orbit.

[0092] It can be understood that in terms of new water body extraction, the present invention addresses the problem that the water body extraction method that only uses scattering information has poor generalization in complex areas with severe speckle noise, and constructs a convolutional neural network to fully utilize the neighborhood spatial texture and scattering information to reduce the impact of speckle noise on water body extraction; uses high-precision DEM data to simulate mountain shadows based on the incident angle and altitude angle when the satellite passes over, and uses post-processing to eliminate mountain shadow areas that are mistakenly classified as water bodies.

[0093] In steps S14-S17, based on the results of farmland element extraction, newly added water bodies caused by flood disasters, and high-precision DEM data, a farmland disaster analysis framework integrating two-dimensional and three-dimensional information is constructed to extract the fully submerged and semi-submerged areas of farmland and output the results of farmland flood disaster analysis. Specifically:

[0094] The spatial distribution result of the cultivated land elements and the newly added water body area are superimposed to obtain the cultivated land fully submerged area;

[0095] Dividing the river data in the DEM data based on the Thiessen polygon method to obtain sub-catchment areas;

[0096] Superimposing the fully submerged cultivated land area and the sub-catchment area, and calculating the average elevation of the fully submerged cultivated land area in each sub-catchment area;

[0097] The elevation of each sub-catchment area is compared respectively, and the cultivated land below the average elevation is determined as the semi-flooded area of ​​cultivated land.

[0098] It should be noted that, firstly, based on the extracted cultivated land elements and the newly added water bodies caused by floods, the overlay analysis outputs the fully submerged cultivated land area, and then based on the high-precision DEM data and river data, the Thiessen polygon method is used to divide the sub-catchment area, and then the average elevation of the fully submerged cultivated land area in each sub-catchment area is obtained by overlaying the fully submerged cultivated land area and the sub-catchment area. Based on the passive inundation model theory, it is assumed that in each sub-catchment area, except for the fully submerged cultivated land area, the cultivated land with an average elevation lower than the fully submerged cultivated land area is a semi-submerged area, and the fully submerged cultivated land area and the semi-submerged cultivated land area are integrated to output the analysis results of the flood damage to cultivated land.

[0099] In one embodiment, the dividing of the river data in the DEM data based on the Thiessen polygon method to obtain the sub-catchment area includes:

[0100] According to the river flow direction grid data, all connected pixel groups belonging to the same drainage basin are obtained, and the pouring points and depression points at the edge of the window are located;

[0101] Identify the catchment area on each of the pour points and construct an initial subcatchment area;

[0102] Extract the DEM data to obtain a water flow direction layer under unfilled conditions, and determine the outlets of each of the initial sub-catchment areas; wherein the point with the lowest ground elevation or the point to which the opposite water flow direction points among all the inclined points with a water flow direction grid change amplitude less than or equal to 45° is the outlet of each sub-catchment area;

[0103] According to the outlets of the initial sub-catchment area, all adjacent outlets are connected to form a triangle, and perpendicular bisectors are drawn for each side of the triangle, and a polygon surrounded by several perpendicular bisectors around each outlet is determined as a sub-catchment area.

[0104] In this embodiment, the specific steps of using the Thiessen polygon method to divide the sub-catchment area are as follows: (1) according to the river flow direction grid data, find all connected pixel groups belonging to the same watershed basin, locate the pouring point (the place where water will pour out of the grid) and the depression point at the edge of the window, and then identify the confluence area on each pouring point to create a sub-catchment area; (2) extract the water flow direction layer under the unfilled depression condition from the DEM of the study area, and set the point with the lowest ground elevation or the point pointing to the opposite water flow direction among several points near the similar water flow direction grid (the change amplitude of each grid is ≤45°) as the outlet of each sub-catchment area; (3) according to the selected sub-catchment area outlet, connect all adjacent outlets into triangles, draw perpendicular bisectors of the sides of these triangles, and the polygon surrounded by several perpendicular bisectors around each outlet is the sub-catchment area.

[0105] In summary, the method for analyzing farmland flood damage provided by the present invention has the following advantages:

[0106] (1) The existing deep learning-based cultivated land feature extraction method can make full use of spectral and spatial information, but the diversity of cultivated land types leads to extremely large intra-class variance, and crop spectra and grassland and woodland spectral information lead to certain misclassification. Compared with the prior art, the present invention introduces elevation and context information when extracting cultivated land to improve classification accuracy, introduces elevation information when cultivated land is generally flat to alleviate the misclassification of cultivated land and mountain grassland and woodland, and constructs a global context information aggregation module to alleviate the intra-class variance of cultivated land and improve classification accuracy.

[0107] (2) The existing methods mainly use the scattering information of radar remote sensing images, which are easily affected by radar image speckle noise when extracting water bodies. Compared with the existing technology, the present invention introduces spatial information to alleviate the image of speckle noise when extracting new water bodies. The present invention constructs a convolutional neural classification network, takes water body marker pixels and neighborhood pixels as network input, and fuses spatial information on the basis of scattering information to reduce the impact of speckle noise on water body extraction.

[0108] (3) The existing method of superimposing the newly added water body on the cultivated land area can only extract the fully submerged area of ​​the cultivated land, but cannot extract the semi-submerged area of ​​the cultivated land. Compared with the existing technology, the present invention can extract the semi-submerged area of ​​the cultivated land. The cultivated land disaster analysis method constructed by the two-dimensional and three-dimensional information fusion of the present invention can realize the extraction of the semi-submerged area of ​​the cultivated land, accurately output the disaster area of ​​the cultivated land, and improve the accuracy of flood disaster monitoring.

[0109] Reference Figure 4 The second embodiment of the present invention provides a device for analyzing farmland flood damage, comprising:

[0110] A data acquisition module is used to acquire multi-source remote sensing data of the flood-affected areas of cultivated land; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images and post-disaster radar images;

[0111] An element extraction module is used to input the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements;

[0112] A water body extraction module, used to extract the pre-disaster radar image and the post-disaster radar image respectively based on a pre-trained convolutional neural network to obtain a newly added water body area;

[0113] A total flooding analysis module is used to superimpose the spatial distribution results of the cultivated land elements and the newly added water body area to obtain the cultivated land total flooding area;

[0114] A river channel division module is used to divide the river channel data in the DEM data based on the Thiessen polygon method to obtain a sub-catchment area;

[0115] An elevation calculation module, used for superimposing the fully flooded cultivated land area and the sub-catchment area, and calculating the average elevation of the fully flooded cultivated land area in each sub-catchment area;

[0116] The semi-flooded analysis module is used to compare the elevations of each of the sub-catchment areas and determine the cultivated land below the average elevation as the semi-flooded cultivated land area.

[0117] It should be noted that the farmland flood damage analysis device provided in the embodiment of the present invention is used to execute all the process steps of the farmland flood damage analysis method of the above embodiment, and the working principles and beneficial effects of the two correspond one to one, so they will not be repeated.

[0118] The embodiment of the present invention further provides a terminal device. The terminal device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a farmland flood damage analysis program. When the processor executes the computer program, the steps in the above-mentioned farmland flood damage analysis method embodiments are implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are realized, such as the semi-flooding analysis module.

[0119] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0120] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a smart tablet. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the above components are merely examples of terminal devices and do not constitute a limitation on the terminal device. The terminal device may include more or fewer components than the above components, or may combine certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.

[0121] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0122] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0123] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0124] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0125] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for analyzing farmland flood damage, characterized in that: include: Acquire multi-source remote sensing data of the flood-affected areas of cultivated land; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images, and post-disaster radar images; Inputting the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements; Based on a pre-trained convolutional neural network, the pre-disaster radar image and the post-disaster radar image are respectively extracted to obtain a newly added water body area; The spatial distribution result of the cultivated land elements and the newly added water body area are superimposed to obtain the cultivated land fully submerged area; According to the river flow direction grid data, all connected pixel groups belonging to the same drainage basin are obtained, and the pouring points and depression points at the edge of the window are located; Identify the catchment area on each of the pour points and construct an initial subcatchment area; Extract the DEM data to obtain a water flow direction layer under unfilled conditions, and determine the outlets of each of the initial sub-catchment areas; wherein the point with the lowest ground elevation or the point to which the opposite water flow direction points among all the inclined points with a water flow direction grid change amplitude less than or equal to 45° is the outlet of each sub-catchment area; According to the outlets of the initial sub-catchment area, all adjacent outlets are connected to form a triangle, and perpendicular bisectors are drawn for each side of the triangle, and a polygon surrounded by several perpendicular bisectors around each outlet is determined as a sub-catchment area; Superimposing the fully submerged cultivated land area and the sub-catchment area, and calculating the average elevation of the fully submerged cultivated land area in each sub-catchment area; The elevation of each sub-catchment area is compared respectively, and the cultivated land below the average elevation is determined as the semi-flooded area of ​​cultivated land.

2. The method for analyzing farmland flood damage according to claim 1, characterized in that: The pre-trained convolutional neural network extracts the pre-disaster radar image and the post-disaster radar image respectively to obtain the newly added water body area, including: Based on a pre-trained convolutional neural network, the pre-disaster radar image and the post-disaster radar image are respectively extracted to obtain an initial water body area; Acquire the satellite's position information when it passes by, and simulate the mountain shadow area when the satellite passes by according to the position information and the DEM data; The initial water body area is screened out according to the mountain shadow area to obtain a newly added water body area.

3. The method for analyzing farmland flood damage according to claim 1, characterized in that: The training process of the farmland feature extraction network includes: Based on the multispectral remote sensing image, a deep spatial-spectral fusion cultivated land element extraction network is constructed; Constructing a labeled data set of the cultivated land remote sensing image based on the multispectral remote sensing image, the DEM data and a preset cultivated land labeled vector map; The farmland feature extraction network is trained according to the labeled data set.

4. The method for analyzing farmland flood damage according to claim 3, characterized in that: The encoder of the farmland feature extraction network includes a global context aggregation module, which includes global context modeling, channel dependency conversion and feature fusion.

5. The method for analyzing farmland flood damage according to claim 1, characterized in that: The training process of the convolutional neural network includes: Annotate water bodies based on the pre-disaster radar images to construct a water body annotation dataset for radar remote sensing images; Convolutional neural network is constructed by taking the image space block composed of the marked pixel as the center and the surrounding neighborhood pixels as input; The convolutional neural network is trained according to the water body annotation dataset.

6. The method for analyzing farmland flood damage according to claim 2, characterized in that: The calculation formula of the hill shadow area includes: Hs=255×((cos(Z)×cos(S))+(sin(Z)×sin(S)×cos(Az_math-As))) Where Z is the zenith angle when the satellite passes by, that is, the angle between the direction of the radar satellite transmitting signal and the local ground plane; S is the slope and aspect data of a certain point on the ground calculated from high-precision DEM data; when Hs is 0, it indicates the mountain shadow area that cannot be covered by the signal when the radar satellite passes by; Az_math is the mathematical azimuth converted from the geographical azimuth Az of the satellite orbit.

7. A device for analyzing farmland flood damage, characterized in that: include: A data acquisition module is used to acquire multi-source remote sensing data of the flood-affected areas of cultivated land; wherein the multi-source remote sensing data includes multi-spectral remote sensing images, DEM data, pre-disaster radar images and post-disaster radar images; An element extraction module is used to input the multispectral remote sensing image and the DEM data into a pre-trained farmland element extraction network to obtain a spatial distribution result of farmland elements; A water body extraction module, used to extract the pre-disaster radar image and the post-disaster radar image respectively based on a pre-trained convolutional neural network to obtain a newly added water body area; A total flooding analysis module is used to superimpose the spatial distribution results of the cultivated land elements and the newly added water body area to obtain the cultivated land total flooding area; The river channel division module is used to obtain all connected pixel groups belonging to the same watershed basin according to the river flow direction grid data, and locate the pouring points and depression points at the edge of the window; identify the confluence area on each pouring point and construct the initial sub-catchment area; extract the DEM data to obtain the water flow direction layer under the condition of not filling the depression, and determine the outlets of each initial sub-catchment area; wherein, the point with the lowest ground elevation or the point pointing to the opposite water flow direction among all the inclined points with a water flow direction grid change amplitude less than or equal to 45° is the outlet of each sub-catchment area; according to the outlet of the initial sub-catchment area, all adjacent outlets are connected into a triangle, and perpendicular bisectors are drawn for each side of the triangle, and the polygon surrounded by several perpendicular bisectors around each outlet is determined as the sub-catchment area; An elevation calculation module, used for superimposing the fully flooded cultivated land area and the sub-catchment area, and calculating the average elevation of the fully flooded cultivated land area in each sub-catchment area; The semi-flooded analysis module is used to compare the elevations of each of the sub-catchment areas and determine the cultivated land below the average elevation as the semi-flooded cultivated land area.

8. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for analyzing farmland flood damage as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for analyzing farmland flood damage according to any one of claims 1 to 6.

Images