A wetland ecological drought identification method based on multi-source data, electronic device and storage medium

Abstract

A method for identifying wetland ecological drought based on multi-source data, an electronic device and a storage medium, belonging to the technical field of intelligent identification of remote sensing data. To solve the problem of accurately identifying wetland ecological drought. The present invention collects wetland image data and meteorological data, constructs wetland image data sets and wetland meteorological data sets; constructs wetland ecological factors, including wetland humidity, rainfall and groundwater level; constructs a deep learning network model, inputs the obtained wetland image data in the wetland image data set into the deep learning network model to extract the water body separation result map of the wetland; obtains the water body distribution map of the wetland based on the obtained wetland water body separation probability map, and solves the change trend of the wetland ecological factors in combination with the obtained wetland meteorological data in the wetland meteorological data set to identify the wetland ecological drought. The present invention provides a scientific basis for wetland protection and ecological environment management, and improves the restoration and protection effect of the wetland ecosystem.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent identification of remote sensing data, and specifically provides a method for identifying wetland ecological drought conditions based on multi-source data, electronic equipment and storage medium. Background Art

[0002] Wetlands are one of the most important ecosystems on Earth. They are transitional zones between land and water, so they have both rich terrestrial and aquatic animal and plant resources, forming a natural gene pool and unique biological environment that cannot be matched by any other single ecosystem. The special soil and climate provide a complex and complete flora and fauna community, which has important ecological and economic value. However, due to climate change and human activities, wetland ecosystems are facing the threat of drought.

[0003] Most of the existing studies have been based on meteorological and hydrological indicators to identify wetland ecological drought and its spatiotemporal dynamics. Since wetlands are special natural complexes formed by the interaction of water and land, they are different from terrestrial and aquatic ecosystems. Periodic hydrological conditions, wetland vegetation and soil are the three major elements that make up wetlands, making them transitional, complex and diverse. Therefore, the existing meteorological and hydrological indicators cannot truly and objectively reflect the complex ecological drought process of wetlands.

[0004] Satellite remote sensing technology has unique advantages in obtaining water information that is complex and changeable in time and space. It can obtain water information accurately and timely due to its short cycle and strong real-time characteristics. Optical and synthetic aperture radar (SAR) remote sensing data have been widely used to detect and map the area and range of surface water. However, due to their data characteristics, optical and microwave imagers can only identify the spatial range or area of ​​water bodies in water body application research. A method that can accurately identify wetland ecological drought conditions is urgently needed. Summary of the invention

[0005] The problem to be solved by the present invention is to accurately identify the ecological drought situation of wetlands, and propose a wetland ecological drought situation identification method based on multi-source data, an electronic device and a storage medium.

[0006] To achieve the above object, the present invention is implemented through the following technical solutions:

[0007] A method for identifying wetland ecological drought conditions based on multi-source data includes the following steps:

[0008] S1. Collect wetland image data and meteorological data, and construct wetland image data set and wetland meteorological data set;

[0009] S2, construct wetland ecological factors, including wetland humidity, rainfall and groundwater level;

[0010] S3, constructing a deep learning network model, inputting the wetland image data in the wetland image data set obtained in step S1 into the deep learning network model to extract the water body separation result map of the wetland;

[0011] S4. Based on the wetland water body separation result map obtained in step S3, a wetland water body distribution map is obtained. Combined with the wetland meteorological data in the wetland meteorological data set obtained in step S1, the change trend of the wetland ecological factors is solved to identify the wetland ecological drought situation.

[0012] Furthermore, the wetland image data in step S1 includes one of satellite remote sensing data, drone remote sensing data, and ground remote sensing data.

[0013] Furthermore, the specific implementation method of step S3 includes the following steps:

[0014] S3.1. Construct a spectral branch of a deep learning network model, wherein the spectral branch includes three 1×1 convolutional layers Conv1×1 for focusing on spectral dimension information; a batch normalization layer and a rectified linear unit BN+ReLU are added after each convolutional layer to improve the nonlinear representation capability of the network, and then a maximum pooling layer MaxPool is added before the second and third convolutional layers to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spectral branch to extract the spectral feature F in the wetland image data. spectral , its calculation expression is:

[0015]

[0016] Among them, Conv 1×1 represents a 1×1 convolutional layer, and channel represents a spectral branch; Represents the output feature map of the lth convolutional layer in the spectral branch; Represents the weight of the lth convolutional layer in the spectral branch; Represents the bias of the lth convolutional layer in the spectral branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer;

[0017] S3.2. Construct a spatial branch of the deep learning network model, wherein the spatial branch includes three 3×3 convolutional layers Conv3×3 for focusing on semantic location information; a batch normalization layer and a rectified linear unit BN+ReLU with an activation function ReLU are added after each convolutional layer to improve the nonlinear representation ability of the network, and then a maximum pooling layer MaxPool is added before the second convolutional layer and the third convolutional layer to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spatial branch to extract the spatial features F in the wetland image data. spatial, its calculation expression is:

[0018]

[0019] Among them, Conv 3×3 represents a 3×3 convolutional layer, and s represents a spatial branch; Represents the output feature map of the lth convolutional layer in the spatial branch; Represents the weight of the lth convolutional layer in the spatial branch; Represents the bias of the lth convolutional layer in the spatial branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer;

[0020] S3.3, build a global spectral spatial attention module to aggregate the spectral features F obtained in step S3.1 spectral and the spatial feature F obtained in step S3.2 spatial Get the integrated feature F G , the calculation expression is:

[0021] F G =F spectral +F spatial

[0022] in, It integrates spectral information and spatial semantic information, where H, W, and C represent the height, width, and number of bands of the integrated feature map, respectively;

[0023] Then two convolutional layers are used to transform F G Divided into features with spectral sensitivity T spectral and the spatially sensitive feature T spatial , the calculation expression is:

[0024] T spectral =conv spectral (F G )

[0025] T spatial =conv spatial (F G )

[0026] Among them, conv spectral and conv spatial The convolution kernel size is 3×3;

[0027] Then the characteristic of spectral sensitivity T spectral , with spatial sensitivity characteristics T spatial The spectral attention map and spatial attention map are obtained by using the spectral and spatial attention modules respectively. The calculation expressions are as follows:

[0028]

[0029] Among them, M spectral and M spatial Represent the spectral attention map and spatial attention map respectively; Att spectral represents the 1-D spectral attention mechanism, Att spatial Represents 2-D spatial attention mechanism; stands for element-wise multiplication;

[0030] S3.4, the spectral attention map and spatial attention map obtained in step S3.3 are respectively compared with the spatial feature F spatial and spectral feature F spectral Adding further highlights the important classification information and suppresses irrelevant information. The calculation expression is:

[0031] F' spectral =F spectral +M spectral

[0032] Among them, F' spectral and F' spatial They represent the spectral feature map after spectral attention enhancement and the spatial feature map after spatial attention enhancement respectively;

[0033] The final spectral feature map and spatial feature map are concatenated in the channel dimension to obtain the spatial-spectral joint feature map, and then the integrated spatial-spectral joint feature map is reshaped into a one-dimensional vector using the Flatten operation, and then input into the fully connected layer for feature fusion, and finally input into the Softmax layer for classification, and the classification probability of each pixel belonging to the water body is calculated. The calculation expression is:

[0034]

[0035] Among them, z i , z j Represent the i-th element and j-th element in the input feature vector respectively;

[0036] The cross entropy loss function Loss is used to update the parameters of the deep learning network model. The calculation expression is:

[0037]

[0038] Among them, B represents the total number of wetland types;

[0039] S3.5. Using the classification probability of each pixel belonging to a water body obtained in step S3.4, classify each pixel to obtain a final wetland water body separation probability map.

[0040] Furthermore, the 1-D spectral attention mechanism Att in step S3.4 spectral The specific implementation method is: use global average pooling AvgPool and global maximum pooling MaxPool to pool the global features of the input spectral sensitivity, and then send the obtained features to a multi-layer perceptron with a hidden layer, and set the hidden layer vector size to Where r represents the attenuation coefficient. Finally, the output of the multilayer perceptron is summed up to get the spectral attention weight. The calculation expression is:

[0041]

[0042] MLP(·)=W1(ReLU(W0(·)))

[0043]

[0044] in, The global feature representing the spectral sensitivity of the global average pooling, represents the global characteristic of spectral sensitivity of global maximum pooling, sigmoid represents the sigmoid activation function; MLP represents multi-layer perceptron, ω spectral Represents the spectral attention weight.

[0045] Furthermore, the 2-D spatial attention mechanism Att in step S3.4 spatial The specific implementation method is: use global average pooling and global maximum pooling to pool the input global features with spatial sensitivity, and then obtain the spatial attention map through standard convolution of the feature maps of the two channels. The calculation expression is:

[0046]

[0047] Among them, f 3×3 The size of the convolution kernel is 3×3, ω spatial Represents the spatial attention weight.

[0048] Furthermore, the specific implementation method of step S4 includes the following steps:

[0049] S4.1. The water area of ​​the wetland is calculated by counting the probability of each pixel belonging to the water body and obtaining the number of pixels classified as water bodies. The calculation expression is:

[0050] S=P water *Res

[0051] Among them, S represents the water area of ​​the wetland, P water is the total number of water body pixels in the wetland water body separation result map, and Res represents the resolution of the input remote sensing data;

[0052] S4.2. Based on the determination of wetland water body classification, the water level is calculated in the water body area using satellite altimetry data. The water level calculation expression is:

[0053] W h =Sal-UD-θ

[0054] Among them, W h represents the reference water level of the ground level surface EGM96, is the actual water level on the ground, Sal represents the distance between the altimeter satellite and the ground, U represents the distance between the ground level surface and the earth's ellipsoid, D represents the distance between the satellite and the water surface, and θ represents the correction parameter;

[0055] The calculation expression of θ is:

[0056] θ=Wtro+Dtro+Iono+Stid+Ptid

[0057] Among them, Wtro and Dtro represent dry and wet tropospheric corrections, respectively, Iono represents ionospheric correction, Stid and Ptid represent solid tide and polar tide corrections, respectively;

[0058] The wetland water body separation result map obtained by S3.4 is combined with the water body area obtained by S4.1 and the water level height information obtained by S4.2 to obtain the final wetland water body distribution map The water body distribution map I contains the wetland water body area information, as well as the area and water level of the water body in the area;

[0059] S4.3. Based on the water distribution map of the wetland obtained in step S4.2, collect the water distribution map of the wetland in the meteorological data set obtained in step S1. Humidity, rainfall and groundwater level data of the region over a long period of time are used to analyze the changing trend of wetland ecological factors, which are represented by y.

[0060] S4.3.1. Arrange the wetland ecological factor data in chronological order, construct data pairs of independent variable time and dependent variable wetland ecological factor, where day represents time; assuming that there is a linear relationship between wetland ecological factor and time day, the calculation expression is:

[0061] y=β0+β1*day+ε

[0062] Among them, β0 and β1 represent coefficients, and ε represents the error term;

[0063] S4.3.2. Calculate the values ​​of regression coefficients β0 and β1 by the least squares method and calculate the mean of the time and the mean of wetland ecological factors The calculation expression is:

[0064]

[0065] The calculation expression of coefficient β1 is:

[0066]

[0067] The calculation expression of coefficient β0 is:

[0068]

[0069] Where n represents the number of data pairs;

[0070] S4.3.3. Calculate the predicted value of the regression equation The calculation expression is:

[0071]

[0072] S4.3.4. Calculation of the coefficient of determination R 2 To evaluate the goodness of fit of the model, R 2 The value of is between 0 and 1. The closer it is to 1, the better the model fit is. The calculation expression is:

[0073]

[0074] Combined with the wetland meteorological data in the wetland meteorological data set obtained in step S1, the variation trend of the wetland ecological factors is solved to identify the wetland ecological drought situation.

[0075] The electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a method for identifying wetland ecological drought conditions based on multi-source data when executing the computer program.

[0076] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the wetland ecological drought condition identification method based on multi-source data.

[0077] Beneficial effects of the present invention:

[0078] The present invention discloses a method for identifying wetland ecological drought based on multi-source data, wherein the satellite sensing data includes surface morphological image data, meteorological data, hydrological data, and geographic information data. By using these multi-source remote sensing data, relevant information of wetlands can be obtained more comprehensively, thereby improving the accuracy of wetland ecological drought identification.

[0079] The present invention discloses a method for identifying wetland ecological drought conditions based on multi-source data, wherein the wetland ecological factors include humidity, rainfall and groundwater level. These factors are important indicators for evaluating wetland drought conditions. By extracting and analyzing these factors, the hydrological conditions of the wetland can be better understood, providing basic data for drought identification.

[0080] The present invention discloses a method for identifying wetland ecological drought conditions based on multi-source data. The fusion of multi-source data can also promote cross-disciplinary research in different disciplines, such as remote sensing technology, meteorology, hydrology, etc., and improve the recognition and understanding of wetland drought conditions.

[0081] The wetland ecological drought identification method based on multi-source data described in the present invention utilizes multi-source remote sensing data to extract and analyze wetland ecological factors, uses image processing algorithms to process satellite image data, and combines meteorological data for analysis to obtain the changing patterns of wetland ecological factors, such as humidity, rainfall and groundwater level, to provide a scientific basis for wetland protection and ecological environment management, help relevant departments formulate corresponding protection and management measures, and improve the restoration and protection effects of wetland ecosystems. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A flowchart of a method for identifying wetland ecological drought conditions based on multi-source data according to the present invention;

[0083] Figure 2 A schematic diagram of the structure of a deep learning network model of a wetland ecological drought condition identification method based on multi-source data according to the present invention;

[0084] Figure 3 This is a schematic diagram of the structure of the global spectral spatial attention module in the deep learning network model of the wetland ecological drought identification method based on multi-source data described in the present invention. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the present invention more clear, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the specific embodiments described are only part of the embodiments of the present invention, rather than all of the specific embodiments. The components of the specific embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations, and the present invention can also have other embodiments.

[0086] Therefore, the following detailed description of the specific embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected specific embodiments of the present invention. Based on the specific embodiments of the present invention, all other specific embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0087] In order to further understand the content, features and effects of the present invention, the following specific implementation methods are given as examples, and the attached Figure 1 -Attached Figure 3 The detailed instructions are as follows: Specific implementation method one:

[0089] A method for identifying wetland ecological drought conditions based on multi-source data includes the following steps:

[0090] S1. Collect wetland image data and meteorological data, and construct wetland image data set and wetland meteorological data set;

[0091] Furthermore, the wetland image data in step S1 includes one of satellite remote sensing data, drone remote sensing data, and ground remote sensing data;

[0092] S2, construct wetland ecological factors, including wetland humidity, rainfall and groundwater level;

[0093] S3, constructing a deep learning network model, inputting the wetland image data in the wetland image data set obtained in step S1 into the deep learning network model to extract the water body separation result map of the wetland;

[0094] Furthermore, the specific implementation method of step S3 includes the following steps:

[0095] S3.1. Construct a spectral branch of a deep learning network model, wherein the spectral branch includes three 1×1 convolutional layers Conv1×1 for focusing on spectral dimension information; a batch normalization layer and a rectified linear unit BN+ReLU are added after each convolutional layer to improve the nonlinear representation capability of the network, and then a maximum pooling layer MaxPool is added before the second and third convolutional layers to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spectral branch to extract the spectral feature F in the wetland image data. spectral , its calculation expression is:

[0096]

[0097] Among them, Conv 1×1 represents a 1×1 convolutional layer, and channel represents a spectral branch; Represents the output feature map of the lth convolutional layer in the spectral branch; Represents the weight of the lth convolutional layer in the spectral branch; Represents the bias of the lth convolutional layer in the spectral branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer;

[0098] S3.2. Construct a spatial branch of the deep learning network model, wherein the spatial branch includes three 3×3 convolutional layers Conv3×3 for focusing on semantic location information; a batch normalization layer and a rectified linear unit BN+ReLU with an activation function ReLU are added after each convolutional layer to improve the nonlinear representation ability of the network, and then a maximum pooling layer MaxPool is added before the second convolutional layer and the third convolutional layer to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spatial branch to extract the spatial features F in the wetland image data. spatial , its calculation expression is:

[0099]

[0100]

[0101] Among them, Conv 3×3 represents a 3×3 convolutional layer, and s represents a spatial branch; Represents the output feature map of the lth convolutional layer in the spatial branch; Represents the weight of the lth convolutional layer in the spatial branch; Represents the bias of the lth convolutional layer in the spatial branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer;

[0102] S3.3, build a global spectral spatial attention module to aggregate the spectral features F obtained in step S3.1 spectral and the spatial feature F obtained in step S3.2 spatial Get the integrated feature F G , the calculation expression is:

[0103] F G =F spectral +F spatial

[0104] in, It integrates spectral information and spatial semantic information, where H, W, and C represent the height, width, and number of bands of the integrated feature map respectively;

[0105] Then two convolutional layers are used to transform F G Divided into features with spectral sensitivity T spectral and the spatially sensitive feature Tspatial , the calculation expression is:

[0106] T spectral =conv spectral (F G )

[0107] T spatial =conv spatial (F G )

[0108] Among them, conv spectral and conv spatial The convolution kernel size is 3×3;

[0109] Then the characteristic of spectral sensitivity T spectral , spatially sensitive features T spatial The spectral attention map and spatial attention map are obtained by using the spectral and spatial attention modules respectively. The calculation expressions are as follows:

[0110]

[0111] Among them, M spectral and M spatial Represent the spectral attention map and spatial attention map respectively; Att spectral represents the 1-D spectral attention mechanism, Att spatial Represents 2-D spatial attention mechanism; stands for element-wise multiplication;

[0112] S3.4, the spectral attention map and spatial attention map obtained in step S3.3 are respectively compared with the spatial feature F spatial and spectral feature F spectral Adding further highlights the important classification information and suppresses irrelevant information. The calculation expression is:

[0113] F' spectral =F spectral +M spectral

[0114] Among them, F' spectral and F' spatial They represent the spectral feature map after spectral attention enhancement and the spatial feature map after spatial attention enhancement, respectively.

[0115] The final spectral feature map and spatial feature map are concatenated in the channel dimension to obtain the spatial-spectral joint feature map, and then the integrated spatial-spectral joint feature map is reshaped into a one-dimensional vector using the Flatten operation. It is then input into the fully connected layer for feature fusion and finally input into the Softmax layer for classification. The classification probability of each pixel belonging to the water body is calculated. The calculation expression is:

[0116]

[0117] Among them, z i , z j Represent the i-th element and j-th element in the input feature vector respectively;

[0118] The cross entropy loss function Loss is used to update the parameters of the deep learning network model. The calculation expression is:

[0119]

[0120] Where B represents the total number of wetland types. Furthermore, the specific implementation method of the 1-D spectral attention mechanism in step S3.4 is as follows: the global features of the input spectral sensitivity are pooled using global average pooling AvgPool and global maximum pooling MaxPool respectively, and then the obtained features are sent to a multi-layer perceptron with a hidden layer, and the hidden layer vector size is set to Where r represents the attenuation coefficient. Finally, the output of the multilayer perceptron is summed up to get the spectral attention weight. The calculation expression is:

[0121]

[0122] MLP(·)=W1(ReLU(W0(·)))

[0123]

[0124] in, The global feature representing the spectral sensitivity of the global average pooling, represents the global characteristic of spectral sensitivity of global maximum pooling, sigmoid represents the sigmoid activation function; MLP represents multi-layer perceptron, ω spectral represents the spectral attention weight;

[0125] Furthermore, the specific implementation method of the 2-D spatial attention mechanism in step S3.4 is: use global average pooling and global maximum pooling to pool the input global features with spatial sensitivity, and then obtain the spatial attention map by standard convolution of the feature maps of the two channels. The calculation expression is:

[0126]

[0127] Among them, f 3×3 The size of the convolution kernel is 3×3, ω spatial represents the spatial attention weight;

[0128] S3.5. Using the classification probability of each pixel belonging to a water body obtained in step S3.4, classify each pixel to obtain the final wetland water body separation probability map.

[0129] S4, obtaining a water distribution map of the wetland based on the water separation result map of the wetland obtained in step S3, solving the change trend of the wetland ecological factors in combination with the wetland meteorological data in the wetland meteorological data set obtained in step S1, and identifying the wetland ecological drought situation;

[0130] Furthermore, the specific implementation method of step S4 includes the following steps:

[0131] S4.1. The water area of ​​the wetland is calculated by counting the probability of each pixel belonging to the water body and obtaining the number of pixels classified as water bodies. The calculation expression is:

[0132] S=P water *Res

[0133] Among them, S represents the water area of ​​the wetland, P water is the total number of water body pixels in the wetland water body separation result map, and Res represents the resolution of the input remote sensing data;

[0134] S4.2. Based on the determination of wetland water body classification, the water level is calculated in the water body area using satellite altimetry data. The water level calculation expression is:

[0135] W h =Sal-UD-θ

[0136] Among them, W h represents the reference water level of the ground level surface EGM96, is the actual water level on the ground, Sal represents the distance between the altimeter satellite and the ground, U represents the distance between the ground level surface and the earth's ellipsoid, D represents the distance between the satellite and the water surface, and θ represents the correction parameter;

[0137] The calculation expression of θ is:

[0138] θ=Wtro+Dtro+Iono+Stid+Ptid

[0139] Among them, Wtro and Dtro represent dry and wet tropospheric corrections, respectively, Iono represents ionospheric correction, Stid and Ptid represent solid tide and polar tide corrections, respectively;

[0140] The wetland water body separation result map obtained by S3.4 is combined with the water body area obtained by S4.1 and the water level height information obtained by S4.2 to obtain the final wetland water body distribution map The water body distribution map I contains the wetland water body area information, as well as the area and water level of the water body in the area.

[0141] S4.3. Based on the water distribution map of the wetland obtained in step S4.2, collect the water distribution map of the wetland in the meteorological data set obtained in step S1. Humidity, rainfall and groundwater level data of the region over a long period of time are used to analyze the changing trend of wetland ecological factors, which are represented by y.

[0142] S4.3.1. Arrange the wetland ecological factor data in chronological order, construct data pairs of independent variable time and dependent variable wetland ecological factor, where day represents time; assuming that there is a linear relationship between wetland ecological factor and time day, the calculation expression is:

[0143] y=β0+β1*day+ε

[0144] Among them, β0 and β1 represent coefficients, and ε represents the error term;

[0145] S4.3.2. Calculate the values ​​of regression coefficients β0 and β1 by the least squares method and calculate the mean of the time and the mean of wetland ecological factors The calculation expression is:

[0146]

[0147] The calculation expression of coefficient β1 is:

[0148]

[0149] The calculation expression of coefficient β0 is:

[0150]

[0151] Where n represents the number of data pairs;

[0152] S4.3.3. Calculate the predicted value of the regression equation The calculation expression is:

[0153]

[0154] S4.3.4. Calculation of the coefficient of determination R 2 To evaluate the goodness of fit of the model, R 2 The value of is between 0 and 1. The closer it is to 1, the better the model fit is. The calculation expression is:

[0155]

[0156] Combined with the wetland meteorological data in the wetland meteorological data set obtained in step S1, the variation trend of the wetland ecological factors is solved to identify the wetland ecological drought situation. Specific implementation method 2:

[0158] The electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a wetland ecological drought identification method based on multi-source data when executing the computer program.

[0159] The computer device of the present invention may be a device including a processor and a memory, such as a single chip microcomputer including a central processing unit, etc. Moreover, the processor is used to implement the steps of the wetland ecological drought identification method based on multi-source data when executing the computer program stored in the memory.

[0160] 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. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0161] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may 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 may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may 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. Specific implementation method three:

[0163] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the wetland ecological drought condition identification method based on multi-source data.

[0164] The computer-readable storage medium of the present invention can be any form of storage medium that can be read by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. The computer-readable storage medium stores a computer program. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above-mentioned method for identifying wetland ecological drought conditions based on multi-source data can be implemented.

[0165] The computer program includes computer program code, which may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer readable medium may 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.

[0166] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0167] Although the present application has been described above with reference to specific embodiments, various modifications may be made thereto and parts thereof may be replaced with equivalents without departing from the scope of the present application. In particular, as long as there is no structural conflict, the various features in the specific embodiments disclosed in the present application may be used in combination with each other in any manner, and the fact that these combinations are not exhaustively described in this specification is only for the sake of omitting space and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for identifying wetland ecological drought conditions based on multi-source data, characterized in that: The steps include: S1. Collect wetland image data and meteorological data, and construct wetland image data set and wetland meteorological data set; S2, construct wetland ecological factors, including wetland humidity, rainfall and groundwater level; S3, constructing a deep learning network model, inputting the wetland image data in the wetland image data set obtained in step S1 into the deep learning network model to extract the water body separation result map of the wetland; The specific implementation method of step S3 includes the following steps: S3.

1. Construct a spectral branch of a deep learning network model, wherein the spectral branch includes three 1×1 convolutional layers Conv1×1 for focusing on spectral dimension information; a batch normalization layer and a rectified linear unit BN+ReLU are added after each convolutional layer to improve the nonlinear representation capability of the network, and then a maximum pooling layer MaxPool is added before the second and third convolutional layers to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spectral branch to extract the spectral feature F in the wetland image data. spectral ; Its calculation expression is Among them, Conv 1×1 represents a 1×1 convolutional layer, and channel represents a spectral branch; Represents the output feature map of the lth convolutional layer in the spectral branch; Represents the weight of the lth convolutional layer in the spectral branch; Represents the bias of the lth convolutional layer in the spectral branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer; S3.

2. Construct a spatial branch of the deep learning network model, wherein the spatial branch includes three 3×3 convolutional layers Conv3×3 for focusing on semantic location information; a batch normalization layer and a rectified linear unit BN+ReLU with an activation function ReLU are added after each convolutional layer to improve the nonlinear representation ability of the network, and then a maximum pooling layer MaxPool is added before the second convolutional layer and the third convolutional layer to reduce the parameters and computational complexity of the network. The wetland image data obtained in step S1 is input into the spatial branch to extract the spatial features F in the wetland image data. spatial , its calculation expression is: Among them, Conv 3×3 represents a 3×3 convolutional layer, and s represents a spatial branch; Represents the output feature map of the lth convolutional layer in the spatial branch; Represents the weight of the lth convolutional layer in the spatial branch; Represents the bias of the lth convolutional layer in the spatial branch; ReLU represents the activation function ReLU; MaxPool represents the maximum pooling layer; S3.3, build a global spectral spatial attention module to aggregate the spectral features F obtained in step S3.1 spectral and the spatial feature F obtained in step S3.2 spatial Get the integrated feature F G , the calculation expression is: F G =F spectral +F spatial in, It integrates spectral information and spatial semantic information, where H, W, and C represent the height, width, and number of bands of the integrated feature map, respectively; Then two convolutional layers are used to transform F G Divided into features T with spectral sensitivity spectral and the spatially sensitive feature T spatial , the calculation expression is: T spectral =conv spectral (F G ) T spatial =conv spatial (F G ) Among them, conv spectral and conv spatial The convolution kernel size is 3×3; Then the characteristic of spectral sensitivity T spectral , spatially sensitive features T spatial The spectral attention map and spatial attention map are obtained by using the spectral and spatial attention modules respectively. The calculation expressions are as follows: Among them, M spectral and M spatial Represent the spectral attention map and spatial attention map respectively; Att spectral represents the 1-D spectral attention mechanism, Att spatial Represents 2-D spatial attention mechanism; stands for element-wise multiplication; S3.4, the spectral attention map and spatial attention map obtained in step S3.3 are respectively compared with the spatial feature F spatial and spectral feature F spectral Adding further highlights the important classification information and suppresses irrelevant information. The calculation expression is: F' spectral =F spectral +M spectral Among them, F' spectral and F' spatial They represent the spectral feature map after spectral attention enhancement and the spatial feature map after spatial attention enhancement respectively; The final spectral feature map and spatial feature map are concatenated in the channel dimension to obtain the spatial-spectral joint feature map, and then the integrated spatial-spectral joint feature map is reshaped into a one-dimensional vector using the Flatten operation, and then input into the fully connected layer for feature fusion, and finally input into the Softmax layer for classification, and the classification probability of each pixel belonging to the water body is calculated. The calculation expression is: Among them, z i , z j Represent the i-th element and j-th element in the input feature vector respectively; the cross entropy loss function Loss is used to update the parameters of the deep learning network model, and the calculation expression is: Among them, B represents the total number of wetland types; S3.5, using the classification probability of each pixel belonging to the water body obtained in step S3.4, classify each pixel to obtain the final wetland water body separation result map; S4. Based on the wetland water body separation result map obtained in step S3, a wetland water body distribution map is obtained. Combined with the wetland meteorological data in the wetland meteorological data set obtained in step S1, the change trend of the wetland ecological factors is solved to identify the wetland ecological drought situation.

2. The wetland ecological drought identification method based on multi-source data according to claim 1 is characterized in that: The wetland image data in step S1 includes one of satellite remote sensing data, drone remote sensing data, and ground remote sensing data.

3. The wetland ecological drought identification method based on multi-source data according to claim 2 is characterized in that: 1-D spectral attention mechanism Att in step S3.4 spectral The specific implementation method is: use global average pooling AvgPool and global maximum pooling MaxPool to pool the global features of the input spectral sensitivity, and then send the obtained features to a multi-layer perceptron with a hidden layer, and set the hidden layer vector size to Where r represents the attenuation coefficient. Finally, the output of the multilayer perceptron is summed up to get the spectral attention weight. The calculation expression is: MLP(·)=W1(ReLU(W0(·))) in, The global feature representing the spectral sensitivity of the global average pooling, represents the global characteristic of spectral sensitivity of global maximum pooling, sigmoid represents the sigmoid activation function; MLP represents multi-layer perceptron, ω spectral Represents the spectral attention weight.

4. The wetland ecological drought identification method based on multi-source data according to claim 3 is characterized in that: 2-D spatial attention mechanism Att in step S3.4 spatial The specific implementation method is: use global average pooling and global maximum pooling to pool the input global features with spatial sensitivity, and then obtain the spatial attention map through standard convolution of the feature maps of the two channels. The calculation expression is: Among them, f 3×3 The size of the convolution kernel is 3×3, ω spatial Represents the spatial attention weight.

5. The wetland ecological drought identification method based on multi-source data according to claim 4 is characterized in that: The specific implementation method of step S4 includes the following steps: S4.

1. The water area of ​​the wetland is calculated by counting the probability of each pixel belonging to the water body and obtaining the number of pixels classified as water bodies. The calculation expression is: S=P water *Res Among them, S represents the water area of ​​the wetland, P water is the total number of water body pixels in the water body separation result map of the wetland, and Res represents the resolution of the input remote sensing data; S4.

2. Based on the determination of wetland water body classification, the water level is calculated in the water body area using satellite altimetry data. The water level calculation expression is: W h =Sal-U-D-θ Among them, W h represents the reference water level of the ground level surface EGM96, is the actual water level on the ground, Sal represents the distance between the altimeter satellite and the ground, U represents the distance between the ground level surface and the earth's ellipsoid, D represents the distance between the satellite and the water surface, and θ represents the correction parameter; The calculation expression of θ is: θ=Wtro+Dtro+Iono+Stid+Ptid Among them, Wtro and Dtro represent dry and wet tropospheric corrections, respectively, Iono represents ionospheric correction, Stid and Ptid represent solid tide and polar tide corrections, respectively; S4.

3. Based on the water distribution map of the wetland obtained in step S4.2, collect the water distribution map of the wetland in the meteorological data set obtained in step S1. Humidity, rainfall and groundwater level data of the region over a long period of time are used to analyze the changing trend of wetland ecological factors, which are represented by y. S4.3.

1. Arrange the wetland ecological factor data in chronological order, construct data pairs of independent variable time and dependent variable wetland ecological factor, where day represents time; assuming that there is a linear relationship between wetland ecological factor and time day, the calculation expression is: y=β0+β1*day+ε Among them, β0 and β1 represent coefficients, and ε represents the error term; S4.3.

2. Calculate the values ​​of regression coefficients β0 and β1 by the least squares method and calculate the mean of the time and the mean of wetland ecological factors The calculation expression is: The calculation expression of coefficient β1 is: The calculation expression of coefficient β0 is: Where n represents the number of data pairs; S4.3.

3. Calculate the predicted value of the regression equation The calculation expression is: S4.3.

4. Calculation of coefficient of determination R 2 To evaluate the goodness of fit of the model, R 2 The value of is between 0 and 1. The closer it is to 1, the better the model fit is. The calculation expression is: Combined with the wetland meteorological data in the wetland meteorological data set obtained in step S1, the variation trend of the wetland ecological factors is solved to identify the wetland ecological drought situation.

6. An electronic device, characterized in that The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of a wetland ecological drought identification method based on multi-source data as described in any one of claims 1 to 5 when executing the computer program.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying wetland ecological drought conditions based on multi-source data as described in any one of claims 1 to 5 is implemented.

Images