Satellite-borne microwave radiometer tropical cyclone wind speed inversion method based on physical constraint deep learning

Abstract

The invention belongs to the technical field of ocean remote sensing, and relates to a spaceborne microwave radiometer tropical cyclone wind speed inversion method based on physical constraint deep learning, comprising the following steps: acquiring tropical cyclone parameter information; extracting multi-band brightness temperature data of the satellite-borne microwave radiometer in a corresponding space-time range, and constructing a sample data set; a deep learning inversion algorithm model with physical constraints is loaded and trained; the physical constraint deep learning inversion algorithm model comprises an input layer, a network layer, a physical layer and an output layer; the network layer is composed of a residual neural network, the physical layer is composed of a D matrix remote sensing statistical formula and provides physical constraints for the whole model, and the D matrix is a nonlinear D matrix; and outputting the tropical cyclone wind speed in the current space-time range. According to the method, the brightness temperature information of the satellite-borne microwave radiometer is obtained by utilizing the tropical cyclone trajectory information, and the network model with physical information is constructed by utilizing the combination of the D matrix equation and the neural network, so that more accurate and high-precision tropical cyclone wind speed information is obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of ocean remote sensing and relates to a tropical cyclone wind speed inversion method using a satellite-borne microwave radiometer based on physical constraint deep learning. Background Art

[0002] A tropical cyclone (TC) is a cyclonic circulation generated over tropical or subtropical oceans, including tropical depressions, tropical storms, severe tropical storms, typhoons, severe typhoons and super typhoons. Those located in the western Pacific Ocean and its adjacent waters are called typhoons, and those located in the Atlantic and eastern Pacific Oceans are called hurricanes. The maximum wind speed of a tropical cyclone can reach over 70 m / s. It carries a huge capacity to transport heat to the polar regions and drive the circulation of the ocean system. However, its extremely high wind speed also brings great disasters, such as floods, landslides, etc., which pose a huge threat to human life and economic losses. Therefore, the high-precision inversion of tropical cyclone wind speed can provide reliable data for human disaster prevention and reduction and for maritime navigation.

[0003] Microwave radiometer wind speed inversion is a method of inferring wind speed by measuring the characteristics of microwave radiation from the sea surface. It has important applications in ocean and meteorological observations, especially in extreme weather conditions such as tropical cyclones. Microwave radiometers passively receive microwave radiation emitted by the earth's surface and use the intensity and spectral characteristics of these radiations to invert surface information, such as sea surface wind speed. Microwave radiometers measure the brightness temperature of sea surface radiation (abbreviated as brightness temperature). This radiation is affected by the roughness of the sea surface, which is mainly caused by wind speed. Therefore, wind speed can be estimated by measuring the brightness temperature of microwave radiation at different frequencies. Generally, higher wind speeds will cause the sea surface to be rougher, which in turn changes the characteristics of microwave radiation.

[0004] Microwave radiometers usually operate in several different frequency bands, and each frequency band has different sensitivities to the sea surface. For example, low-frequency microwaves (such as 6-10 GHz) mainly reflect changes in sea surface temperature and humidity, while high-frequency microwaves (such as 30-40 GHz) are more susceptible to sea surface roughness and rainfall. Therefore, in wind speed inversion, it is necessary to combine data from multiple frequency bands for comprehensive analysis.

[0005] Radiative transfer models are used to link the observed brightness temperature to the sea surface wind speed. These models need to take into account various influencing factors, such as atmospheric absorption, scattering, sea surface reflection, etc. Through accurate radiative transfer calculations, more accurate sea surface wind speed can be inverted.

[0006] In tropical cyclones, microwave radiometers can provide key wind speed data to help monitor cyclone intensity and path. Due to the presence of strong winds and precipitation in cyclones, higher requirements are placed on the inversion of microwave signals. This requires more complex algorithms to distinguish the contribution of wind speed and precipitation to microwave brightness temperature.

[0007] The challenges faced by microwave radiometer wind speed inversion technology include reducing the impact of atmospheric transmittance attenuation under high rainfall, reducing the impact of increased sea surface emissivity caused by wave breaking under high wind speed, and improving wind speed accuracy under extreme wind speed conditions. In order to solve these problems, it is imperative to design a comprehensive algorithm that retains the radiation transfer model and introduces a more advanced inversion algorithm.

[0008] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0009] The purpose of the present invention is to solve the technical problems existing in the prior art such as the lack of physical mechanism in deep learning algorithms and the low accuracy of the existing traditional microwave radiometer high wind speed inversion algorithms. A tropical cyclone wind speed inversion method based on physical constraint deep learning of satellite-borne microwave radiometer is proposed. The brightness temperature information of the satellite-borne microwave radiometer is obtained by using the tropical cyclone trajectory information, and a network model with physical information is constructed by combining the D matrix equation with the neural network to obtain more accurate and high-precision tropical cyclone wind speed information.

[0010] The technical solution of the present invention is: A method for inverting tropical cyclone wind speed using a spaceborne microwave radiometer based on physical constraint deep learning, characterized in that it comprises the following steps: S101, obtaining tropical cyclone parameter information, wherein the tropical cyclone parameter information includes: historical and current movement tracks of the tropical cyclone center, including the center longitude and latitude, and time; S102, extracting multi-band brightness temperature data of satellite-borne microwave radiometers in the corresponding time and space range, and constructing a sample data set using brightness temperature data extracted from historical tropical cyclone parameter information and airborne tropical cyclone wind speed data; S103, loading a deep learning inversion algorithm model with physical constraints, and using a sample data set to train the deep learning inversion algorithm model with physical constraints to obtain a deep learning wind speed inversion algorithm model with physical constraints; The physical constraint deep learning inversion algorithm model includes: an input layer, a network layer, a physical layer, and an output layer; the network layer is composed of a residual neural network, and the residual network has 3 fully connected layers and 2 residual layers; the physical layer is composed of a D matrix remote sensing statistical formula (the difference between the D matrix remote sensing statistical formula and the traditional D matrix statistical formula lies in its coefficients: the coefficients in the traditional D matrix formula are fixed, while the coefficients of the D matrix remote sensing statistical formula are dynamic coefficients generated by the residual network, which can change dynamically according to the brightness temperature of multiple bands), adding physical constraints to the entire network training process and giving physical interpretability to deep learning, wherein the D matrix is ​​a nonlinear D matrix; S104. After extracting the microwave radiometer multi-band brightness temperature data of the current tropical cyclone, the brightness temperature data to be inverted is input into a deep learning wind speed inversion algorithm model with physical constraints, and the tropical cyclone wind speed in the current time and space range is output.

[0011] Furthermore, in step S102, based on the moving track of historical tropical cyclones, brightness temperature data of the satellite-borne microwave radiometer scanning on the track is extracted, and the extracted brightness temperature data and airborne tropical cyclone wind speed data are used to construct a sample data set required for the model.

[0012] Furthermore, the movement trajectory includes the center longitude and latitude and time; Based on the tropical cyclone center position and track time information, satellite-borne microwave radiometer brightness temperature data, and airborne tropical cyclone wind speed data, a sample data set is constructed, including: Based on the central parameter information of historical tropical cyclones (central longitude and latitude and track time information), the multi-band brightness temperature data within the range of ±1000 km in longitude and latitude and the range of ±0.5 hour in the central track time of the tropical cyclone (from the Advanced Microwave Scanning Radiometer-2, AMSR2) were determined as the input data for constructing the sample dataset; the output data of the sample dataset is the more accurate historical airborne tropical cyclone wind speed data (from the Stepped Frequency Microwave Radiometer, SMFR); the two types of data are matched according to the time and space method to obtain the sample dataset used.

[0013] Furthermore, in step S103, the physical constraint deep learning inversion algorithm model is trained using the sample data set, including: (1) Divide the sample data set into a training set and a test set according to the preset ratio; (2) using the training set to train the physical constraint deep learning inversion algorithm model to obtain an intermediate physical constraint deep learning inversion algorithm model; (3) Using the test set, the intermediate physical constraint deep learning inversion algorithm model is tested, and the intermediate physical constraint deep learning inversion algorithm model that has completed the test is determined as a deep learning wind speed inversion algorithm model with physical constraints.

[0014] Furthermore, the training set includes a plurality of sample data, and the physical constraint deep learning inversion algorithm model is trained using the training set, including: (2.1) Network layer calculation, including full connection calculation and residual calculation. The residual connection calculation formula is: (4) in, y k(l-1) It is l-1 Tier k The output value of a neuron, f r represents the Relu activation function, k= 1… K , l= 1… L ; y kl , w,b Respectively expressed in l Tier k The output, weight and bias of each neuron; x Indicates that from l-1 Layer input to l The value of the layer; the last layer of the network l=5, k=0~10, the result y kl , the result is also the D matrix coefficient required by the physical layer, expressed as: (5); (2.2) Physical layer calculation: The wind speed is calculated by using the D matrix equation through the D matrix coefficient and the multi-band brightness temperature data of the input layer. Here, the D matrix equation plays the role of physically constraining the entire model. The calculation formula is: (6); in, (7) (8); (2.3) Loss calculation: Loss calculation is performed based on the wind speed calculated by the D matrix and the wind speed in the training set; (2.4) Reverse callback step: perform reverse callback on the parameters of the physical constraint deep learning inversion algorithm model based on the inversion loss, and repeat the network layer calculation step, physical layer calculation step and loss calculation step until the inversion loss no longer decreases or the number of repeated executions reaches 3000 times, then the loop is stopped. The model obtained after the stop is determined as the intermediate physical constraint deep learning inversion algorithm model.

[0015] Furthermore, the input data of the test set is input into the intermediate physical constraint deep learning inversion algorithm model to obtain its wind speed data, and the error is calculated with the output data of the test set (i.e., output wind speed); When the error value is within the set range, a deep learning wind speed inversion algorithm model with physical constraints is obtained; when the error value exceeds the set range, the network parameters are readjusted and the model is trained.

[0016] Furthermore, if the error value of the test set differs from the error value of the training set by 1 m / s, the final deep learning wind speed inversion algorithm model with physical constraints is obtained; if the difference between the two error values ​​exceeds 1 m / s, the model is retrained.

[0017] Furthermore, in step S103, the input layer needs to jump to the physical layer while connecting to the network layer.

[0018] Furthermore, in step S104, the brightness temperature data of the current tropical cyclone is input into a deep learning wind speed inversion algorithm model with physical constraints to obtain high-precision tropical cyclone wind speed and matrix coefficients of the D matrix in the model.

[0019] In the current tropical cyclone wind speed inversion process, the matrix coefficients of the D matrix generated by the model can be used for physical characteristic analysis. This coefficient constrains the impact of factors such as rainfall in extreme weather on wind speed inversion.

[0020] Beneficial effects of the present invention: (1) The present invention provides a tropical cyclone wind speed inversion method based on a residual network with physical constraints. By using the trajectory information of the tropical cyclone to obtain the brightness temperature data of the satellite-borne microwave radiometer, and combining the D matrix equation with the neural network to construct a network model with physical information, the influence of the changes in sea surface brightness temperature received by the satellite-borne microwave radiometer caused by the changes in sea surface emissivity and atmospheric transmittance caused by rainfall and wave breaking can be effectively reduced, thereby obtaining more accurate tropical cyclone wind speed information.

[0021] (2) The dynamic D matrix coefficients in the present invention fully explain the physical information of the ocean atmosphere used by deep learning in the process of sea surface wind speed inversion. The D matrix formula also fully constrains the training of deep learning, effectively solving the problem of lack of physical support in the process of deep learning in the inversion of ocean parameters, and providing high-quality and reliable wind speed products suitable for extreme weather. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of the tropical cyclone wind speed inversion method using a spaceborne microwave radiometer based on physical constraint deep learning provided by the present invention; Figure 2 Schematic diagram of the physical constraint deep learning wind speed inversion algorithm provided by the present invention. DETAILED DESCRIPTION

[0023] 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.

[0024] In order to further understand the present invention, the present invention will be further described in conjunction with the accompanying drawings and embodiments.

[0025] The present invention provides Figure 1 An embodiment of a method for inverting tropical cyclone wind speed using a spaceborne microwave radiometer based on physical constraint deep learning is shown.

[0026] In such Figure 1 The steps shown in the flowcharts shown may be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in an order different from that shown here.

[0027] This embodiment includes the following steps: S101, obtaining tropical cyclone parameter information including the longitude and latitude of the tropical cyclone center and time; The sample tropical cyclone parameter information data are derived from the CMA best track dataset, the China Meteorological Administration Typhoon Network, and the International Climate Management Best Tracking Archive (IBTrACS).

[0028] S102, extract the multi-band brightness temperature data of the spaceborne microwave radiometer in the corresponding time and space range. The data comes from the Advanced Microwave Scanning Radiometer-2 (AMSR2) carried by the Global Change Observation Mission 1st Water (GCOM-W1). The multi-band brightness temperature data are shown in Table 1. In the table, Res06 represents 6.9 GHz resolution, AMSR2 L1R represents L1R-level data of AMSR2, AMSR2 L3 represents L3-level data of AMSR2, and K represents Kelvin temperature unit (Kelvins, K); the corresponding time and space range is the spatial range of the central longitude and latitude ±1000 km, and the time range is ±0.5 hours.

[0029] Table 1 Multi-band brightness temperature dataset

[0030] Then, the dataset was constructed, with AMSR2 brightness temperature data as the input sample of the dataset, and the output sample as the airborne tropical cyclone wind speed data, which comes from the Stepped Frequency Microwave Radiometer (SMFR data). The two types of data were matched in time and space, with a spatial range of ±25 km and a spatial range of ±0.5 hours. Other required data for the dataset are shown in Table 1.

[0031] S103, construct a physical constraint deep learning inversion algorithm model, and train and load it.

[0032] The physical constraint deep learning inversion algorithm model is trained using the sample data set to obtain a deep learning wind speed inversion algorithm model with physical constraints based on microwave radiometer; this model automatically adjusts the wind speed inversion contribution of each brightness temperature frequency band to achieve the purpose of physical constraints and physical analysis.

[0033] In an embodiment of the present invention, the physical constraint deep learning inversion algorithm model includes: an input layer, a network layer, a physical layer and an output layer, wherein the network layer is composed of a residual neural network, and the residual network has 3 fully connected layers and 2 residual layers; the physical layer is composed of a D matrix remote sensing statistical formula. Unlike the traditional D matrix formula, the coefficients in the traditional D matrix formula are fixed, while the coefficients of the D matrix remote sensing statistical formula are dynamic coefficients generated by the residual network, which can change dynamically according to the multi-band brightness temperature. Its construction form is as follows Figure 2 shown.

[0034] In the embodiment of the present invention, step S103 includes the following steps: (1) Divide the sample data set into a training set and a test set according to the preset ratio of 7:3; (2) Using the training set, the physical constraint deep learning inversion algorithm model is trained to obtain an intermediate physical constraint deep learning inversion algorithm model; (3) Using the test set, the intermediate physical constraint deep learning inversion algorithm model is tested, and the intermediate physical constraint deep learning inversion algorithm model that has completed the test is determined as the deep learning wind speed inversion algorithm model with physical constraints.

[0035] Specifically, the training set contains multiple sample data, and step (2) also includes the following network construction and training steps: The input layer and output layer are the brightness temperature of AMSR2 L1R and SSW_sfmr of SFMR. The input involves 10 frequency bands (TB6.9h, TB6.9v, TB10.7h, TB10.7v, TB18.7h, TB18.7v, TB23.8h, TB23.8v, TB36.5h and TB36.5v), which together form an N ∗ 10 data matrix: (1) Where N represents the number of data sets. i= 1… N , h Indicates the input brightness temperature of this frequency band H polarization mode, v Indicates the data set input brightness temperature of this frequency band V polarization mode. The output forms an N*1 data matrix: (2) Fully connected calculation steps: Input data enters the first hidden layer for calculation: (3) in, k= 1… K , l= 1… L ; K represents the total number of neurons in the network layer l, L represents the total number of layers in the network, y kl Indicated in l Tier k The output of a neuron; w Indicated in l Tier k The weights of the neurons; b Indicated in l Tier k The bias of each neuron; x Indicates that from l -1 layer input to l The value of f t Represents the Tanh activation function.

[0037] Residual connection calculation steps: The first hidden layer of the second layer is a residual layer. The result of the previous layer is jump-connected to this layer. The calculation method is: (4) in, y k(l-1) It is l-1 Tier k The output value of a neuron, f r represents the Relu activation function, k= 1… K , l= 1… L; y kl , w,b Respectively expressed in l Tier k The output, weight and bias of each neuron; x Indicates that from l-1 Layer input to l The calculation method of other layers of the network layer is similar to formula (3) and formula (4). The output result of the special hidden layer 5 (hidden layer 5) is passed to the physical layer. The result is the coefficient of the D matrix required by the physical layer, that is: (5) Contribution matrix calculation steps for wind speed: Using the above D coefficient and the brightness temperature data of the input layer, the required wind speed can be calculated through the D matrix equation. The calculation formula is as follows: (6) in, (7) (8) Formula (5) is a nonlinear D matrix equation and is also the data-driven function of the physical layer. Log(TBs) is the Log equation of brightness temperature, and Ds is the D coefficient obtained at the network layer, which is the contribution of brightness temperature in each frequency band to wind speed.

[0038] Obviously, this process involves satisfying two conditions: first, the network layer (such as Figure 2 The solution of the D matrix equation (shown in Figure 1) must satisfy the coefficients of the D matrix equation of the randomly selected points in the region, representing the physical information part of the network. Secondly, the solution to the D matrix equation must satisfy the output conditions of the wind speed. The process model automatically adjusts the output of the D coefficient according to the data, so that the low-frequency brightness temperature provides a higher contribution in the wind speed inversion process, and the high-frequency brightness temperature provides a lower contribution in the wind speed inversion process, especially during heavy rainfall, thereby achieving the purpose of reducing the impact of rainfall.

[0039] The loss calculation step uses the wind speed calculated by the D matrix and the label wind speed (i.e., the output wind speed of the data set) to calculate its loss. The loss function is the root mean square error (MSE): All the above steps are executed repeatedly. When the loss remains unchanged or the number of repetitions reaches 3000, the loop is stopped and the obtained model is regarded as the intermediate physical constraint deep learning inversion algorithm model. Then proceed to step (3).

[0040] S104, after obtaining the brightness temperature data of the tropical cyclone to be inverted, the brightness temperature data of the current tropical cyclone is input into the deep learning wind speed inversion algorithm model with physical constraints that has completed the test, and the wind speed of the ocean tropical cyclone is inverted in real time to obtain the wind speed information of the current tropical cyclone. And through the D coefficient in the inversion process, physical analysis is performed in the inversion process.

[0041] The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still modify the technical solutions described in the above embodiments, or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, modification, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for tropical cyclone wind speed inversion using spaceborne microwave radiometer based on physical constraint deep learning, characterized in that: The steps include: S101, obtaining tropical cyclone parameter information, including historical and current tropical cyclone center longitude and latitude and time; S102, extracting multi-band brightness temperature data of satellite-borne microwave radiometers in the corresponding time and space range, and constructing a sample data set using brightness temperature data extracted from historical tropical cyclone parameter information and airborne tropical cyclone wind speed data; S103, loading a deep learning inversion algorithm model with physical constraints, and using a sample data set to train the deep learning inversion algorithm model with physical constraints to obtain a deep learning wind speed inversion algorithm model with physical constraints; The physical constraint deep learning inversion algorithm model includes: an input layer, a network layer, a physical layer, and an output layer; the network layer is composed of a residual neural network, the physical layer is composed of a D matrix remote sensing statistical formula, and the D matrix is ​​a nonlinear D matrix; S104: After extracting the microwave radiometer multi-band brightness temperature data of the current tropical cyclone, input it into a deep learning wind speed inversion algorithm model with physical constraints to output the tropical cyclone wind speed in the current time and space range.

2. According to the method of claim 1, the method for tropical cyclone wind speed inversion based on physical constraint deep learning using satellite-borne microwave radiometer is characterized in that: In step S102, according to the moving track of historical tropical cyclones, the brightness temperature data of the satellite-borne microwave radiometer scanning on the track is extracted, and the sample data set required by the model is constructed using the extracted brightness temperature data and the airborne tropical cyclone wind speed data.

3. The method for tropical cyclone wind speed inversion based on physical constraint deep learning of spaceborne microwave radiometer according to claim 2 is characterized in that: The movement trajectory includes the center longitude and latitude and time; Based on the central longitude and latitude and time information of historical tropical cyclones, the multi-band brightness temperature data within the range of ±1000 km of longitude and latitude and the range of ±0.5 hour of the central track time of the tropical cyclone were determined as the input data for constructing the sample dataset; the historical airborne tropical cyclone wind speed data were used as the output data of the sample dataset; the two types of data were matched according to the time-space matching method to obtain the sample dataset used.

4. The method for tropical cyclone wind speed inversion using a spaceborne microwave radiometer based on physical constraint deep learning according to claim 1, characterized in that: In step S103, the physical constraint deep learning inversion algorithm model is trained using the sample data set, including: (1) Divide the sample data set into a training set and a test set according to the preset ratio; (2) using the training set to train the physical constraint deep learning inversion algorithm model to obtain an intermediate physical constraint deep learning inversion algorithm model; (3) Using the test set, the intermediate physical constraint deep learning inversion algorithm model is tested, and the intermediate physical constraint deep learning inversion algorithm model that has completed the test is determined as a deep learning wind speed inversion algorithm model with physical constraints.

5. The method for tropical cyclone wind speed inversion based on physical constraint deep learning of spaceborne microwave radiometer according to claim 4 is characterized in that: The training set is used to train the physical constraint deep learning inversion algorithm model, including: (2.1) Network layer calculation, including full connection calculation and residual calculation. The residual connection calculation formula is: （4） in, y k(l-1) It is l-1 Tier k The output value of a neuron, f r represents the Relu activation function, k= 1… K , l= 1… L ; y kl , w,b Respectively expressed in l Tier k The output, weight and bias of each neuron; x Indicates that from l-1 Layer input to l The value of the layer; the last layer of the network l=5, k=0~10, the result y kl , the result is also the D matrix coefficient required by the physical layer, expressed as: (5); (2.2) Physical layer calculation: The wind speed is calculated using the D matrix equation through the obtained D matrix coefficients and the multi-band brightness temperature data of the input layer. Here, the D matrix equation plays the role of physically constraining the entire model. The calculation formula is: (6); in, (7) （8）； (2.3) Loss calculation: Loss calculation is performed based on the wind speed calculated by the D matrix and the output data of the training set; (2.4) Based on the inversion loss, the parameters of the physical constraint deep learning inversion algorithm model are reversed and the network layer calculation steps, physical layer calculation steps and loss calculation steps are repeatedly executed until the inversion loss no longer decreases or the number of repeated executions reaches 3000 times, and the loop is stopped. The model obtained after the stop is determined as the intermediate physical constraint deep learning inversion algorithm model.

6. The method for tropical cyclone wind speed inversion using a spaceborne microwave radiometer based on physical constraint deep learning according to claim 4, characterized in that: Input the input data of the test set into the intermediate physical constraint deep learning inversion algorithm model to obtain its wind speed data, and calculate the error with the output data of the test set; When the error value is within the set range, a deep learning wind speed inversion algorithm model with physical constraints is obtained; when the error value exceeds the set range, the network parameters are readjusted and the model is trained.

7. The method for tropical cyclone wind speed inversion using a spaceborne microwave radiometer based on physical constraint deep learning according to claim 6, characterized in that: If the error value of the test set differs from that of the training set by 1 m / s, the final deep learning wind speed inversion algorithm model with physical constraints is obtained; if the difference between the two error values ​​exceeds 1 m / s, the model is retrained.

8. The method for tropical cyclone wind speed inversion using a spaceborne microwave radiometer based on physical constraint deep learning according to claim 1, characterized in that: In step S103, the input layer needs to jump to the physical layer while connecting to the network layer.

9. The method for tropical cyclone wind speed inversion using a spaceborne microwave radiometer based on physical constraint deep learning according to claim 1, characterized in that: In step S104, the brightness temperature data of the current tropical cyclone, i.e., the brightness temperature data to be inverted, is input into a deep learning wind speed inversion algorithm model with physical constraints to obtain high-precision tropical cyclone wind speed and matrix coefficients of the D matrix in the model. The matrix coefficients of the D matrix are used for physical characteristic analysis during the current tropical cyclone wind speed inversion process.

Images