CN111639746B - GNSS-R sea surface wind speed inversion method and system based on CNN neural network - Google Patents
GNSS-R sea surface wind speed inversion method and system based on CNN neural network Download PDFInfo
- Publication number
- CN111639746B CN111639746B CN202010406706.3A CN202010406706A CN111639746B CN 111639746 B CN111639746 B CN 111639746B CN 202010406706 A CN202010406706 A CN 202010406706A CN 111639746 B CN111639746 B CN 111639746B
- Authority
- CN
- China
- Prior art keywords
- wind speed
- neural network
- layer
- cnn neural
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a GNSS-R sea surface wind speed inversion method and system based on a CNN neural network, wherein the method comprises the following steps: inputting a DDM diagram to be tested into a pre-trained sea surface wind speed inversion model, and outputting a corresponding inversion wind speed; the sea surface wind speed inversion model is a CNN neural network. According to the invention, the CNN neural network is utilized to invert the GNSS-R sea surface wind speed, the model is simple, the modeling time and inversion time are shortened, and the inversion precision is further improved; the CNN neural network fully utilizes the physical quantity related to the wind speed in the DDM to perform characteristic learning, reduces the calculated quantity and shortens the time consumption under the condition of guaranteeing the inversion precision, and has the characteristics of simple and quick model, high result precision and the like.
Description
Technical Field
The invention relates to the field of atmospheric science research, in particular to a GNSS-R sea surface wind speed inversion method and system based on a CNN neural network.
Background
Sea surface wind speed is a vital physical parameter in ocean state information, and can be detected through a GNSS-R satellite remote sensing technology at present. Because the GNSS-R technology has the characteristics of high global coverage, high space-time resolution and the like, high-quality sea surface wind speed detection data can be obtained. The current GNSS-R wind speed inversion method mainly comprises the following two steps:
waveform matching method: firstly, extracting system state information according to actual measurement data, then regenerating a theoretical model simulation waveform, finally carrying out normalization processing to obtain a theoretical waveform diagram, and establishing a simulation waveform database based on a large number of theoretical waveform diagrams; and generating a waveform diagram to be measured from the measured data during inversion, and carrying out noise reduction and normalization processing. And matching the waveform diagram to be detected with a theoretical waveform diagram in a database, wherein the wind speed corresponding to the successfully matched theoretical waveform diagram is the sea surface wind speed of the data to be detected. However, this method has the disadvantage of large calculation and extremely time-consuming construction of a fine database.
Empirical function method: through experience summarization of a large amount of measured data, one or two physical parameters with high correlation with sea surface wind speed are selected from the DDM, regression linear fitting is carried out, and therefore function mapping between the DDM and the sea surface wind speed is established to obtain wind speed. The sea surface wind speed is often not the result of one or two parameters alone, and therefore the accuracy of the method is affected by disregarding other physical parameters.
Disclosure of Invention
The invention aims to overcome the defects of the two methods, and mainly comprises a waveform matching method, large calculated amount and long inversion time; the empirical function method has poor inversion accuracy. Compared with a waveform matching method, the GNSS-R sea surface wind speed inversion method based on the CNN is free from establishing a huge simulation database, and compared with an empirical function method, the GNSS-R sea surface wind speed inversion method based on the CNN can establish a plurality of relations between observed quantity and sea surface wind speed and fully utilize physical quantity related to wind speed inversion, so that the inversion time can be further shortened, and the inversion precision can be improved.
To achieve the above objective, embodiment 1 of the present invention provides a GNSS-R sea surface wind speed inversion method based on a CNN neural network, the method comprising:
inputting a DDM diagram to be tested into a pre-trained sea surface wind speed inversion model, and outputting a corresponding inversion wind speed; the sea surface wind speed inversion model is a CNN neural network.
As an improvement of the above method, the CNN neural network sequentially includes: the input layer is a DDM graph, and the node number of the input layer is 2560; the number of the output characteristic graphs of the first convolution layer C1 is 1, and the number of the neurons is 1024; the number of the output characteristic graphs of the second convolution layer is 32, 64, 128 and 256 in sequence, the number of neurons is 512, and the convolution kernels of the two convolution layers are 3*3 in size; the domain sizes of the first pooling layer and the second pooling layer are 2 x 2; the number of neurons of the full connection layer between the convolution layer and the output layer is 16; the number of output layer nodes is 1, and the output of the node is sea surface wind speed; the activation function of the CNN neural network is a ReLU function; the loss function is an MSE function, and the evaluation index is Root Mean Square Error (RMSE).
As an improvement of the above method, the method further comprises: the training step of the CNN neural network specifically comprises the following steps:
selecting a plurality of groups of GNSS-R data and ECMWF analysis field data, and performing space-time matching to obtain an original sample set, wherein each group of samples consists of a DDM graph and corresponding wind speed;
preprocessing an original sample set and cutting the original sample set into a training set and a testing set;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish a mapping relation with wind speed;
and testing the trained CNN neural network by using the test set data.
As an improvement of the method, the original sample set is preprocessed and segmented into a training set and a testing set; the method specifically comprises the following steps:
screening the original sample set data based on longitude and latitude, wind speed and signal to noise ratio;
preprocessing the screened original sample set data based on a sampling algorithm and a normalization algorithm;
the preprocessed original sample set is segmented into a training set and a testing set according to the proportion of 7:3.
As an improvement of the method, the CNN neural network is continuously trained through training set data, so that the network can continuously capture data features in the DDM graph and establish a mapping relation with wind speed, and the method specifically comprises the following steps:
after weights among neurons in the CNN neural network and threshold values in each neuron are initialized, training set data are transmitted into a first convolution layer from an input layer to carry out convolution calculation, a feature map output after convolution is processed through a nonlinear activation function, and then the data space dimension is reduced through a first pooling layer to obtain a feature matrix; then, a second convolution layer and a second pooling layer are input to carry out convolution and pooling operations, all data features obtained by convolution are combined through a full connection layer, the normalized data features are transmitted to an output layer, an error between the output wind speed and the real wind speed is calculated through a loss function, and the result is reversely transmitted from the output layer by layer, and weights of the convolution layers and the pooling layers are adjusted;
and repeating the forward propagation and the backward propagation processes, gradually reducing the result of the loss function until the result is in a desired error range or reaches the set training times, and completing the training of the CNN neural network to realize the nonlinear mapping from the DDM graph to the sea surface wind speed.
The embodiment 2 of the invention provides a GNSS-R sea surface wind speed inversion system based on a CNN neural network, which comprises: a trained sea surface wind speed inversion model and a wind speed inversion module; the sea surface wind speed inversion model is a CNN neural network;
the wind speed inversion module is used for inputting the DDM diagram to be tested into a trained sea surface wind speed inversion model and outputting the corresponding inversion wind speed.
As an improvement of the above system, the CNN neural network sequentially includes: the input layer is a DDM graph, and the node number of the input layer is 2560; the number of the output characteristic graphs of the first convolution layer C1 is 1, and the number of the neurons is 1024; the number of the output characteristic graphs of the second convolution layer is 32, 64, 128 and 256 in sequence, the number of neurons is 512, and the convolution kernels of the two convolution layers are 3*3 in size; the domain sizes of the first pooling layer and the second pooling layer are 2 x 2; the number of neurons of the full connection layer between the convolution layer and the output layer is 16; the number of output layer nodes is 1, and the output of the node is sea surface wind speed; the activation function of the CNN neural network is a ReLU function; the loss function is an MSE function, and the evaluation index is Root Mean Square Error (RMSE).
As an improvement of the above system, the training step of the CNN neural network specifically includes:
selecting a plurality of groups of GNSS-R data and ECMWF analysis field data, and performing space-time matching to obtain an original sample set, wherein each group of samples consists of a DDM graph and corresponding wind speed;
preprocessing an original sample set and cutting the original sample set into a training set and a testing set;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish a mapping relation with wind speed;
and testing the trained CNN neural network by using the test set data.
As an improvement of the system, the original sample set is preprocessed and segmented into a training set and a testing set; the method specifically comprises the following steps:
screening the original sample set data based on longitude and latitude, wind speed and signal to noise ratio;
preprocessing the screened original sample set data based on a sampling algorithm and a normalization algorithm;
the preprocessed original sample set is segmented into a training set and a testing set according to the proportion of 7:3.
As an improvement of the above system, the training of the CNN neural network by the training set data continuously enables the network to continuously capture the data features in the DDM graph and establish the mapping relationship with the wind speed, which specifically includes:
after weights among neurons in the CNN neural network and threshold values in each neuron are initialized, training set data are transmitted into a first convolution layer from an input layer to carry out convolution calculation, a feature map output after convolution is processed through a nonlinear activation function, and then the data space dimension is reduced through a first pooling layer to obtain a feature matrix; then, a second convolution layer and a second pooling layer are input to carry out convolution and pooling operations, all data features obtained by convolution are combined through a full connection layer, the normalized data features are transmitted to an output layer, an error between the output wind speed and the real wind speed is calculated through a loss function, and the result is reversely transmitted from the output layer by layer, and weights of the convolution layers and the pooling layers are adjusted;
and repeating the forward propagation and the backward propagation processes, gradually reducing the result of the loss function until the result is in a desired error range or reaches the set training times, and completing the training of the CNN neural network to realize the nonlinear mapping from the DDM graph to the sea surface wind speed.
The invention has the advantages that:
1. compared with a waveform matching method, the GNSS-R sea surface wind speed inversion method based on the CNN (Convolutional Neural Network) convolutional neural network does not need to establish a huge simulation database, and compared with an empirical function method, the GNSS-R sea surface wind speed inversion method based on the CNN (Convolutional Neural Network) convolutional neural network can establish a plurality of relations between observed quantity and sea surface wind speed and fully utilize physical quantity related to wind speed inversion, so that the inversion time can be further shortened and the inversion precision can be improved;
2. according to the invention, the CNN neural network is utilized to invert the GNSS-R sea surface wind speed, the model is simple, the modeling time and inversion time are shortened, and the inversion precision is further improved;
3. the invention fully utilizes the physical quantity related to wind speed in the DDM to perform characteristic learning based on the CNN neural network, reduces the calculated quantity and shortens the time consumption under the condition of guaranteeing the inversion precision, and has the characteristics of simple and quick model, high result precision and the like;
4. the method can be relatively and efficiently based on the advantage of inverting the sea surface wind speed of the CNN neural network, and can meet the requirement of carrying out the atmospheric study related to the sea surface wind field by utilizing a large amount of GNSS-R satellite data.
Drawings
FIG. 1 is a flow chart of a GNSS-R sea surface wind speed inversion method based on a CNN neural network;
fig. 2 is a schematic diagram of a CNN neural network of a sea surface wind speed inversion model.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and specific embodiments.
Example 1
As shown in fig. 1, embodiment 1 of the present invention provides a GNSS-R sea surface wind speed inversion method based on a CNN neural network, which mainly includes the following steps:
the first step: the original data sample set is constructed. Performing space-time matching on GNSS-R data and ECMWF data to form an original sample set;
and selecting a large amount of GNSS-R data and ECMWF analysis field data for space-time matching to obtain an original sample set, wherein each set of samples consists of a DDM graph and corresponding wind speed information.
And a second step of: a training set and a testing set are generated. Preprocessing an original sample set and cutting the original sample set into a training set and a testing set;
in order to avoid data anomaly and noise interference, the longitude and latitude, wind speed and signal-to-noise ratio (SNR) of the data need to be screened; after screening, the problems of nonuniform distribution and inconsistent dimension of the data set are solved based on a sampling algorithm and a normalization algorithm, and the preprocessed sample set is segmented into a training set and a testing set according to a ratio of 7:3.
And a third step of: and constructing a CNN neural network. Constructing a CNN neural network model taking a DDM graph as input and a wind speed as output;
firstly, determining an activation function of a network, then respectively determining the number of network layers, the number of network nodes at each layer and the optimal iteration times of a model according to continuous experiments, then selecting an evaluation index of the model, and finally constructing a CNN neural network model taking a DDM graph as input and a wind speed as output.
Fourth step: CNN neural network training. Continuously training a CNN model through training set data, so that the model can continuously capture data features in a DDM graph and establish a mapping relation with wind speed;
after the weights among the neurons in the network and the threshold value in each neuron are initialized, training set data are transmitted into a first convolution layer from an input layer to carry out convolution calculation, the feature map output after convolution is processed through a nonlinear activation function, and then the data space dimension is reduced through a pooling layer to obtain a feature matrix. Repeating convolution and pooling operations, combining all data features obtained by convolution through a full connection layer in a network after all convolution operations are completed, finally reducing sensitivity of output of feature mapping to translation and other transformations through normalization, transmitting a result to an output layer, calculating an error between an output wind speed and a real wind speed through a loss function, reversely transmitting the result layer by layer from the output layer, and adjusting weights of all hidden layers. Repeating the forward propagation and backward propagation processes, gradually reducing the result of the loss function until the result is in the expected error range of the model or reaches the training times set by the model, predicting the wind speed to approach the real wind speed, and considering that model training is completed, thereby realizing nonlinear mapping from the DDM graph to the sea surface wind speed.
Fifth step: and (5) model testing. Verifying the accuracy and reliability of the trained model through the test set;
and testing the trained model by using test set data, if error results of a plurality of test sets are similar, the model has robustness, judging the accuracy and reliability of the inversion result of the model according to whether the RMSE obtained by the model is within 2m/s, and if the conditions are met, obtaining the CNN neural network sea surface wind speed inversion model.
Sixth step: and (5) inverting the data. Inputting the DDM diagram to be measured into a CNN neural network sea surface wind speed inversion model to obtain the corresponding inversion wind speed.
The CNN-based neural network is adopted to invert the GNSS-R sea surface wind speed by utilizing the observation data of the SGR-ReSI GNSS-R receiver on the TDS-1 satellite. The TDS-1 satellite was launched in 2014, operated on an orbit with a height of 635km and an inclination of 98.4 °, using SSTL-150 platforms, with 8 payloads on the satellite periodically alternating, with SGR-desi operating times of 1-2 days in one operating cycle during which the reflected signals from the GNSS satellites on the earth were received and processed. The TDS-1 satellite generates a Doppler (Delay-Doppler Map, DDM) diagram by incoherently accumulating attribute data attached to the reflected signals, wherein the size of the DDM diagram is 128 Delay pixels multiplied by 20 Doppler pixels, the Doppler resolution is 500Hz, and the Delay resolution is 0.25chips.
As shown in fig. 1, the method comprises the following six steps:
first step, constructing original data samples: the GNSS-R data is space-time matched with the ECMWF data. In the embodiment, TDS-1 satellite data of 2018 2-10 months and ECMWF analysis field data are used, an original sample set is obtained by matching according to time, longitude and latitude, the data volume of the whole sample set reaches more than twenty hundred thousand samples, the whole ocean area is basically covered, and the wind speed range is 0-20 m/s.
Generating a training set and a testing set: and screening the original sample set, removing sample points of sea ice areas with the latitude of the north and south hemispheres higher than 55 degrees, and simultaneously, only leaving data with the signal-to-noise ratio (SNR) higher than 3 in the range of 3-18 m/s in order to avoid noise affecting the wind speed inversion accuracy. The screened sample data are about 37 ten thousand, the wind speed is mostly 3-10m/s, the wind speed is uniformly distributed by adopting a mixed sampling algorithm, the wind speed of a sample set is mapped to a range of 0-1 by adopting a normalization algorithm to ensure the consistent dimension, and finally the sample set obtained by processing is segmented into training sets and test sets according to a ratio of 7:3, wherein the number of the training sets is 73500, and the number of the test sets is 31500.
Thirdly, constructing a CNN neural network: firstly, determining that an activation function of a network is a ReLU function, and then determining that the network consists of an input layer i, a convolution layer C1, a pooling layer S1, a convolution layer C2, a pooling layer S2, a full connection layer and an output layer according to continuous experiments. The number of nodes of the input layer is 2560, the number of output feature images of the convolution layer C1 is 1, the number of neurons is 1024, the number of output feature images of the convolution layer C2 is 32, 64, 128 and 256 in sequence, the number of neurons is 512, and the sizes of convolution kernels arranged on the convolution layer are 3*3; the domain sizes of the pooling layers S1 and S2 are 2 x 2, the number of neurons of a full-connection layer between the convolution layer and the output layer is 16, the number of nodes of the output layer is 1, the MSE function is selected by the loss function, then the evaluation index of the selected model is Root Mean Square Error (RMSE), and a CNN neural network model taking a DDM graph as input and wind speed as output is built.
Fourth step CNN neural network training: after the weights among neurons in a network and the threshold value in each neuron are initialized, an input layer receives a DDM image as input data, the DDM image is transmitted into a first convolution layer, all feature spaces of the DDM image are acted by each convolution kernel of the layer, weighted summation calculation is carried out on pixels of each sensing area and corresponding connection weights, an obtained result plus an offset value is a calculation result of the convolution kernel, the feature image output after convolution is processed through a ReLU nonlinear activation function, and data space dimension is reduced through maximum pooling (Max pooling) to obtain a feature matrix. Repeating convolution and pooling operations, combining all data features obtained by convolution through a full connection layer in a network after all convolution operations are completed, finally reducing sensitivity of output of feature mapping to translation and other transformations through normalization, transmitting a result to an output layer, calculating an error between an output wind speed and a real wind speed through a loss function MSE, reversely transmitting the result layer by layer from the output layer, and adjusting weights of hidden layers. And repeating the forward propagation and the backward propagation processes, wherein the calculation error of the neural network model is continuously reduced, so that the predicted value gradually approaches to the real wind speed. When the calculation error does not decrease along with the increase of training times, an optimal model can be obtained, and the model training is considered to be completed, so that the nonlinear mapping from the DDM graph to the sea surface wind speed is realized.
Fifth step, model test: and testing the trained model by using test set data, if error results of a plurality of test sets are similar, the model has robustness, and judging the accuracy and reliability of the inversion result of the model according to whether the RMSE obtained by the model is within 2m/s, wherein the error range in the example is 1.5-1.8m/s, and the accuracy requirement is met, so that the CNN neural network sea surface wind speed inversion model is obtained.
Sixth step, inversion of data: inputting the DDM diagram to be detected into a CNN neural network sea surface wind speed inversion model, outputting to obtain a corresponding inversion wind speed, and obtaining the RMSE of 1.52m/s.
Example 2
The embodiment 2 of the invention provides a GNSS-R sea surface wind speed inversion system based on a CNN neural network, which comprises: a trained sea surface wind speed inversion model and a wind speed inversion module; the sea surface wind speed inversion model is a CNN neural network;
the wind speed inversion module is used for inputting the DDM diagram to be tested into a trained sea surface wind speed inversion model and outputting the corresponding inversion wind speed.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention and are not limiting. Although the present invention has been described in detail with reference to the embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the appended claims.
Claims (4)
1. A GNSS-R sea surface wind speed inversion method based on a CNN neural network, the method comprising:
inputting a DDM diagram to be tested into a pre-trained sea surface wind speed inversion model, and outputting a corresponding inversion wind speed; the sea surface wind speed inversion model is a CNN neural network;
the CNN neural network sequentially comprises: the input layer is a DDM graph, and the node number of the input layer is 2560; the number of the output characteristic graphs of the first convolution layer C1 is 1, and the number of the neurons is 1024; the number of the output characteristic graphs of the second convolution layer is 32, 64, 128 and 256 in sequence, the number of the neurons is 512, and the convolution kernels arranged on the two convolution layers are all of the same sizeThe method comprises the steps of carrying out a first treatment on the surface of the First, theThe domain sizes of the first pooling layer and the second pooling layer are bothThe method comprises the steps of carrying out a first treatment on the surface of the The number of neurons of the full connection layer between the convolution layer and the output layer is 16; the number of output layer nodes is 1, and the output of the node is sea surface wind speed; the activation function of the CNN neural network is a ReLU function; the loss function is an MSE function, and the evaluation index is Root Mean Square Error (RMSE);
the method further comprises the steps of: the training step of the CNN neural network specifically comprises the following steps:
selecting a plurality of groups of GNSS-R data and ECMWF analysis field data, and performing space-time matching to obtain an original sample set, wherein each group of samples consists of a DDM graph and corresponding wind speed;
preprocessing an original sample set and cutting the original sample set into a training set and a testing set;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish a mapping relation with wind speed;
testing the trained CNN neural network by using test set data;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish the mapping relation with the wind speed, and the CNN neural network specifically comprises the following steps:
after weights among neurons in the CNN neural network and threshold values in each neuron are initialized, training set data are transmitted into a first convolution layer from an input layer to carry out convolution calculation, a feature map output after convolution is processed through a nonlinear activation function, and then the data space dimension is reduced through a first pooling layer to obtain a feature matrix; then, a second convolution layer and a second pooling layer are input to carry out convolution and pooling operations, all data features obtained by convolution are combined through a full connection layer, the normalized data features are transmitted to an output layer, an error between the output wind speed and the real wind speed is calculated through a loss function, the error is reversely transmitted from the output layer by layer, and weights of the convolution layers and the pooling layers are adjusted;
and repeating the forward propagation and the backward propagation processes, gradually reducing the result of the loss function until the result is in a desired error range or reaches the set training times, and completing the training of the CNN neural network to realize the nonlinear mapping from the DDM graph to the sea surface wind speed.
2. The CNN neural network-based GNSS-R sea surface wind speed inversion method according to claim 1, wherein the original sample set is preprocessed and segmented into a training set and a testing set; the method specifically comprises the following steps:
screening the original sample set data based on longitude and latitude, wind speed and signal to noise ratio;
preprocessing the screened original sample set data based on a sampling algorithm and a normalization algorithm;
the preprocessed original sample set is segmented into a training set and a testing set according to the proportion of 7:3.
3. A GNSS-R sea surface wind speed inversion system based on a CNN neural network, the system comprising: a trained sea surface wind speed inversion model and a wind speed inversion module; the sea surface wind speed inversion model is a CNN neural network;
the wind speed inversion module is used for inputting the DDM diagram to be tested into a trained sea surface wind speed inversion model and outputting a corresponding inversion wind speed;
the CNN neural network sequentially comprises: the input layer is a DDM graph, and the node number of the input layer is 2560; the number of the output characteristic graphs of the first convolution layer C1 is 1, and the number of the neurons is 1024; the number of the output characteristic graphs of the second convolution layer is 32, 64, 128 and 256 in sequence, the number of the neurons is 512, and the convolution kernels arranged on the two convolution layers are all of the same sizeThe method comprises the steps of carrying out a first treatment on the surface of the The domain sizes of the first pooling layer and the second pooling layer are bothThe method comprises the steps of carrying out a first treatment on the surface of the Convolutional layerThe number of neurons of the full connection layer between the output layers is 16; the number of output layer nodes is 1, and the output of the node is sea surface wind speed; the activation function of the CNN neural network is a ReLU function; the loss function is an MSE function, and the evaluation index is Root Mean Square Error (RMSE);
the training step of the CNN neural network specifically comprises the following steps:
selecting a plurality of groups of GNSS-R data and ECMWF analysis field data, and performing space-time matching to obtain an original sample set, wherein each group of samples consists of a DDM graph and corresponding wind speed;
preprocessing an original sample set and cutting the original sample set into a training set and a testing set;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish a mapping relation with wind speed;
testing the trained CNN neural network by using test set data;
the CNN neural network is continuously trained through the training set data, so that the network can continuously capture the data characteristics in the DDM graph and establish the mapping relation with the wind speed, and the CNN neural network specifically comprises the following steps:
after weights among neurons in the CNN neural network and threshold values in each neuron are initialized, training set data are transmitted into a first convolution layer from an input layer to carry out convolution calculation, a feature map output after convolution is processed through a nonlinear activation function, and then the data space dimension is reduced through a first pooling layer to obtain a feature matrix; then, a second convolution layer and a second pooling layer are input to carry out convolution and pooling operations, all data features obtained by convolution are combined through a full connection layer, the normalized data features are transmitted to an output layer, an error between the output wind speed and the real wind speed is calculated through a loss function, the error is reversely transmitted from the output layer by layer, and weights of the convolution layers and the pooling layers are adjusted;
and repeating the forward propagation and the backward propagation processes, gradually reducing the result of the loss function until the result is in a desired error range or reaches the set training times, and completing the training of the CNN neural network to realize the nonlinear mapping from the DDM graph to the sea surface wind speed.
4. The CNN neural network-based GNSS-R sea surface wind speed inversion system according to claim 3, wherein said preprocessing and slicing the original sample set into a training set and a test set; the method specifically comprises the following steps:
screening the original sample set data based on longitude and latitude, wind speed and signal to noise ratio;
preprocessing the screened original sample set data based on a sampling algorithm and a normalization algorithm;
the preprocessed original sample set is segmented into a training set and a testing set according to the proportion of 7:3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010406706.3A CN111639746B (en) | 2020-05-14 | 2020-05-14 | GNSS-R sea surface wind speed inversion method and system based on CNN neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010406706.3A CN111639746B (en) | 2020-05-14 | 2020-05-14 | GNSS-R sea surface wind speed inversion method and system based on CNN neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111639746A CN111639746A (en) | 2020-09-08 |
| CN111639746B true CN111639746B (en) | 2023-10-20 |
Family
ID=72330077
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010406706.3A Active CN111639746B (en) | 2020-05-14 | 2020-05-14 | GNSS-R sea surface wind speed inversion method and system based on CNN neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111639746B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112180369B (en) * | 2020-09-23 | 2021-09-10 | 中国人民解放军国防科技大学 | Depth learning-based sea surface wind speed inversion method for one-dimensional synthetic aperture radiometer |
| CN112433233B (en) * | 2020-11-19 | 2023-01-03 | 武汉大学 | GNSS-R sea surface wind speed inversion method and system based on particle swarm optimization |
| CN112733386B (en) * | 2021-01-25 | 2021-08-17 | 吉林大学 | Multi-component reactive solute transport parameter inversion method based on deep neural network |
| CN113687352A (en) * | 2021-08-05 | 2021-11-23 | 南京航空航天大学 | Inversion method for down-track interferometric synthetic aperture radar sea surface flow field |
| CN114912365B (en) * | 2022-06-02 | 2024-06-25 | 重庆大学 | Wind pressure short-term prediction method and device based on PARALLEL CNN-GRU and computer equipment |
| CN115410086A (en) * | 2022-08-26 | 2022-11-29 | 南方海洋科学与工程广东省实验室(广州) | Water quality inversion method, device and equipment based on remote sensing image |
| CN116068595B (en) * | 2023-04-06 | 2023-06-30 | 极诺星空(北京)科技有限公司 | Sea surface wind speed inversion method and device, electronic equipment and medium |
| CN117798654B (en) * | 2024-02-29 | 2024-05-03 | 山西漳电科学技术研究院(有限公司) | Intelligent adjusting system for center of steam turbine shafting |
| CN117969881B (en) * | 2024-03-29 | 2024-06-25 | 中国石油大学(华东) | Method, device and medium for inverting typhoon sea state and sea surface wind speed based on remote sensing data |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
-
2020
- 2020-05-14 CN CN202010406706.3A patent/CN111639746B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780906A (en) * | 2016-12-28 | 2017-05-31 | 北京品恩科技股份有限公司 | A kind of testimony of a witness unification recognition methods and system based on depth convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 吴崇 ; 严颂华 ; 杨永立 ; 卜方玲 ; 陈泽强.基于时延-多普勒图像的海洋表面风速反演方法.科技通报.2019,第第35卷卷(第第9期期),第23-27页. * |
| 吴崇，严颂华，杨永立，卜方玲，陈泽强.基于时延－多普勒图像的海洋表面风速反演方法.科技通报.2019,第第35卷卷(第第9期期),第23-27页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111639746A (en) | 2020-09-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111639746B (en) | GNSS-R sea surface wind speed inversion method and system based on CNN neural network | |
| CN111639747B (en) | GNSS-R sea surface wind speed inversion method and system based on BP neural network | |
| US5581490A (en) | Contact management model assessment system for contact tracking in the presence of model uncertainty and noise | |
| CN112180369B (en) | Depth learning-based sea surface wind speed inversion method for one-dimensional synthetic aperture radiometer | |
| CN113189014B (en) | Ozone concentration estimation method integrating satellite remote sensing and ground monitoring data | |
| CN114758230A (en) | Underground target body classification and identification method based on attention mechanism | |
| CN111983619B (en) | Underwater acoustic target forward scattering acoustic disturbance positioning method based on transfer learning | |
| CN114894819B (en) | Satellite-borne GNSS-R soil humidity inversion method | |
| CN112990082B (en) | Detection and identification method of underwater sound pulse signal | |
| CN115690632A (en) | Water environment monitoring method for inland river water body | |
| CN114048681A (en) | DOA estimation method, system, storage medium and device based on self-selection neural network | |
| CN113253248B (en) | Small sample vertical array target distance estimation method based on transfer learning | |
| CN114219998A (en) | Sonar image real-time detection method based on target detection neural network | |
| Champenois et al. | Machine learning framework for the real-time reconstruction of regional 4D ocean temperature fields from historical reanalysis data and real-time satellite and buoy surface measurements | |
| Liu et al. | GNSS-R global sea surface wind speed retrieval based on deep learning | |
| CN117671504A (en) | Marine wind power identification method and system based on yolo algorithm | |
| Tafazoli et al. | Performance improvement of GPS GDOP approximation using recurrent wavelet neural network | |
| TANG et al. | Application of LVQ neural network combined with the genetic algorithm in acoustic seafloor classification | |
| CN113109795B (en) | Deep sea direct sound zone target depth estimation method based on deep neural network | |
| CN114529815A (en) | Deep learning-based traffic detection method, device, medium and terminal | |
| Xie et al. | A Convolution Neural Network-based Method for Sea Ice Remote Sensing using GNSS-R Data | |
| CN114063063A (en) | Geological disaster monitoring method based on synthetic aperture radar and point-like sensor | |
| Champenois et al. | Real-time reconstruction of 3D ocean temperature fields from reanalysis data and satellite and buoy surface measurements | |
| CN113534085B (en) | Sea surface wind speed and effective wave height joint inversion method of interference imaging altimeter | |
| CN113657416B (en) | Deep sea sound source ranging method and system based on improved deep neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |