A method for correcting omnidirectional wave height spectrum of satellite-borne spectrometer

Abstract

The present invention discloses a method for correcting the omnidirectional wave height spectrum of a satellite-borne spectrometer, which belongs to the field of microwave remote sensing. The method adopts a deep learning network and designs a loss function according to the spectral characteristics of the SWIM omnidirectional wave height spectrum to train the deep learning network. Considering that the omnidirectional wave height spectrum is also affected by residual speckle noise at high wavenumbers (the wavenumber range greater than the peak wavenumber), and has poor consistency with the reference spectrum at high wavenumbers, thereby affecting the performance of the spectral parameters, a penalty term is added to the loss function to increase the penalty at high wavenumbers: the part with a wavenumber greater than twice the peak wavenumber is regarded as the high wavenumber part that needs to be punished more severely, which can ensure that the corrected SWIM omnidirectional wave height spectrum has good consistency with the buoy at high wavenumbers.

Description

Technical Field

[0001] The present invention belongs to the field of microwave remote sensing, and more specifically, relates to a method for correcting omnidirectional wave height spectrum of a satellite-borne spectrometer. Background Art

[0002] The ocean is a treasure trove of natural resources awaiting exploration, development, and utilization. As a direct manifestation of ocean motion, the detection of ocean wave spectra has significant scientific and applied value. Current technologies for detecting ocean waves primarily include on-site detection and remote sensing.

[0003] With the advancement of satellite remote sensing technology, the China-France Oceanography Satellite (CFOSAT) has been successfully launched. The Surface Waves Investigation and Monitoring (SWIM) instrument aboard CFOSAT is the world's first satellite-borne spectrometer, capable of detecting ocean waves in the 70-500m wavelength range. The resulting wave spectrum data is of great value for marine science research and applications.

[0004] Current research on the performance of the satellite-borne spectrometer SWIM ocean wave spectrum product has revealed poor agreement between the omnidirectional wave height spectra measured by SWIM and those obtained from buoys in low sea conditions. This is primarily due to the presence of spurious peaks at low wavenumbers in the SWIM omnidirectional wave height spectrum. Furthermore, compared to the buoy's omnidirectional wave height spectrum, the peak wavenumber of the SWIM omnidirectional spectrum is shifted to the left, resulting in a lower peak value. This is due to the surfboard effect. These factors, along with other unknown nonlinear effects, can degrade the performance of the SWIM ocean wave spectrum product, and these spectral discrepancies are currently difficult to correct analytically.

[0005] In recent years, artificial neural networks have been introduced into the field of ocean parameter correction, such as significant wave height and mean wave period. However, there has been no research on the correction of SWIM wave spectrum products themselves. Summary of the Invention

[0006] In view of the above defects or improvement needs of the prior art, the present invention provides a method for correcting the omnidirectional wave height spectrum of a satellite-borne spectrometer to improve the performance of the satellite-borne spectrometer's ocean wave spectrum product.

[0007] To achieve the above object, according to a first aspect of the present invention, a method for training an omnidirectional wave height spectrum correction model for a satellite-borne spectrometer is provided, comprising:

[0008] S1, pre-processing the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy to make them correspond one to one;

[0009] S2, construct a loss function to minimize the difference between the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the corresponding buoy, and train the satellite-borne spectrometer omnidirectional wave height spectrum correction model;

[0010] The satellite-borne spectrometer omnidirectional wave height spectrum correction model is a neural network model; the loss function is: M is the batch size of training data; N is the number of wavenumber points in the unified wavenumber domain; k j is the jth wavenumber point in the unified wavenumber domain; is the peak wave number of the omnidirectional wave height spectrum measured by the buoy corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; is the SWIM omnidirectional wave height spectrum of the i-th sample after correction by the neural network model; is the omnidirectional wave height spectrum measured by the buoy corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; η * i is the dimensionless variance corresponding to the i-th spectrum to be corrected in a batch of samples; w ij It is a penalty coefficient based on the sea conditions and the relationship between the wave number and the peak wave number. a1 and a2 are the penalty coefficients of the high wave number part of the SWIM omnidirectional wave height spectrum in pure wind waves and pure swell waves, respectively.

[0011] According to the second aspect of the present invention, a method for correcting the omnidirectional wave height spectrum of a satellite-borne spectrometer is provided, in which the SWIM omnidirectional wave height spectrum to be corrected is input into the satellite-borne spectrometer omnidirectional wave height spectrum correction model trained using the training method described in the first aspect to obtain the corrected omnidirectional wave height spectrum.

[0012] According to a third aspect of the present invention, there is provided a training system for an omnidirectional wave height spectrum correction model of a satellite-borne spectrometer, comprising: a computer-readable storage medium and a processor;

[0013] The computer-readable storage medium is used to store executable instructions;

[0014] The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the training method as described in the first aspect.

[0015] According to a fourth aspect of the present invention, there is provided a satellite-borne spectrometer omnidirectional wave height spectrum correction system comprising: a computer-readable storage medium and a processor;

[0016] The computer-readable storage medium is used to store executable instructions;

[0017] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the correction method as described in the second invention.

[0018] According to a fifth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the training method as described in the first aspect, or execute the correction method as described in the second aspect.

[0019] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0020] 1. The method provided by the present invention adopts a deep learning network and designs a loss function according to the spectral characteristics of the SWIM omnidirectional wave height spectrum to train the deep learning network; wherein, considering that the omnidirectional wave height spectrum is also affected by residual speckle noise at high wavenumbers (the wavenumber range greater than the peak wavenumber), the consistency with the reference spectrum at high wavenumbers is poor, which in turn affects the performance of the spectral parameters. Therefore, a penalty term is added to the loss function to increase the penalty at high wavenumbers: the part with a wavenumber greater than twice the peak wavenumber is regarded as the high wavenumber part that needs to be punished more severely, which can ensure that the corrected SWIM omnidirectional wave height spectrum is consistent with the buoy at high wavenumbers.

[0021] 2. The method provided by the present invention combines the characteristics of SWIM omnidirectional wave height spectrum to improve the U-Net network model to obtain the SWIM omnidirectional wave height spectrum correction network model. The difference between it and the U-Net network model is that: (1) considering that the SWIM omnidirectional wave height spectrum data has negative spectral values, the SWIM omnidirectional wave height spectrum correction network model uses the leakyReLU activation function in the encoder part to learn the features of negative spectral values ​​for subsequent correction; compared with the use of the ReLU activation function in the encoder part, the model can more effectively correct the negative spectral values; through the jump connection between the encoder and the decoder, the feature information before and after correction of different dimensions is extracted and fused, so that the network model can effectively correct the SWIM omnidirectional wave height spectrum; (2) The SWIM omnidirectional wave height spectrum correction network model adds a batch normalization (BN) layer after the convolution layer and the transposed convolution layer to solve the problem of internal covariate drift, thereby accelerating the learning speed of the network model and reducing the sensitivity of the network model to the initial learning rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A schematic diagram of a method for correcting omnidirectional wave height spectrum of a satellite-borne spectrometer provided in an embodiment of the present invention;

[0023] Figure 2 A schematic diagram of the SWIM omnidirectional wave height spectrum correction network model structure provided by an embodiment of the present invention;

[0024] Figure 3(a) to (c) are the comparison results of the average omnidirectional wave height spectra of SWIM and NDBC buoys before and after correction under the condition of 6-degree beam single wind and wave, U≤12m / s, and Ω at different values. Figure 3 (d) to (f) are the comparison results of the average omnidirectional wave height spectra of SWIM and NDBC buoys before and after correction under single wind and wave conditions with U>12m / s and Ω at different values; where U is the wind speed and Ω is the inverse wave age;

[0025] Figure 4 (a) to (c) are respectively the conditions of pure surge with 6-degree beam, H s =1, 1.4m, k p Comparison of the average omnidirectional wave height spectra of SWIM and NDBC buoys before and after correction at different values. Figure 4 (d) to (f) in the figure are respectively the H s =1.8, 2.2m, k p Comparison of the average omnidirectional wave height spectra of SWIM and NDBC buoys before and after correction at different values; s is the effective wave height, k p is the peak wave number. DETAILED DESCRIPTION

[0026] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0027] Deep learning has strong nonlinear fitting capabilities, and considering that the omnidirectional wave height spectrum has certain waveform characteristics, based on this, the present invention adopts a deep learning network and designs a loss function according to the spectral characteristics of the SWIM omnidirectional wave height spectrum to calibrate the parameters of the deep learning network.

[0028] An embodiment of the present invention provides a method for training an omnidirectional wave height spectrum correction model for a satellite-borne spectrometer, comprising:

[0029] S1, preprocessing the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy to make them correspond one to one.

[0030] Specifically, the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy are preprocessed. After the data preprocessing process, each SWIM omnidirectional wave height spectrum sample input into the network has a corresponding buoy omnidirectional wave height spectrum as a reference spectrum, that is, the correction benchmark.

[0031] Furthermore, the preprocessing includes:

[0032] The SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy are successively unified in wavenumber domain, equalized and standardized.

[0033] Specifically, the wavenumber ranges of the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy are obtained respectively, and the wavenumber domains of the two are unified (that is, the common wavenumber range of the two); the sample sizes corresponding to the pure surge wave and pure wind wave sea conditions are balanced; the SWIM omnidirectional wave height spectrum to be corrected and the omnidirectional wave height spectrum measured by the buoy after the balanced operation are standardized to obtain the standardized spectrum of the SWIM omnidirectional wave height spectrum to be corrected and the reference spectrum of the buoy omnidirectional wave height spectrum; the reference spectrum of the buoy omnidirectional wave height spectrum is used as the benchmark for correcting the SWIM omnidirectional wave height spectrum to be corrected.

[0034] S2, construct a loss function to minimize the difference between the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the corresponding buoy, and train the satellite-borne spectrometer omnidirectional wave height spectrum correction model;

[0035] Specifically, the loss function of the regression layer is calculated using Mean Square Error (MSE), and L2 regularization is used to prevent overfitting.

[0036] The omnidirectional wave height spectrum is also affected by residual speckle noise at high wavenumbers (the wavenumber range greater than the peak wavenumber), and has poor consistency with the reference spectrum at high wavenumbers, which in turn affects the performance of the spectrum parameters. Considering that the omnidirectional wave height spectrum at high wavenumbers has a smaller value, the present invention increases the penalty at high wavenumbers by adding a penalty term to the loss function; the consistency between the spectrum to be corrected and the reference spectrum at wavenumbers greater than twice the peak wavenumber is poor, so this part is penalized more, and the part with wavenumbers greater than twice the peak wavenumber is regarded as the high wavenumber part that needs to be penalized more. The weighted loss function with increased penalty for the high wavenumber part is expressed as follows:

[0037]

[0038] in,

[0039]

[0040] M is the batch size of training data; N is the number of wavenumber points in the unified wavenumber domain; k j is the jth wavenumber point in the unified wavenumber domain; is the peak wave number of the reference spectrum corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; is the SWIM omnidirectional wave height spectrum of the i-th sample after correction by the neural network model; is the reference spectrum corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; η * i is the dimensionless variance corresponding to the i-th spectrum to be corrected in a batch of samples; w ij It is a penalty coefficient based on the sea conditions and the relationship between the wave number and the peak wave number. a1 and a2 are the penalty coefficients of the high wave number part of the SWIM omnidirectional wave height spectrum of pure wind waves and pure surge waves, respectively.

[0041] Preferably, the neural network model is a BU-Net network model; the BU-Net network model is obtained by processing a U-Net network model;

[0042] The processing includes: adding a batch normalization layer after each convolutional layer in the encoder, each convolutional layer in the decoder and the transposed convolutional layer of the U-Net network model, and modifying the activation function of the encoder: using the leakyReLU activation function as the activation function of the encoder.

[0043] Specifically, the present invention combines the characteristics of the SWIM omnidirectional wave height spectrum to improve the U-Net network model and obtain the BU-Net network model, which is used as the SWIM omnidirectional wave height spectrum correction network model. The difference between it and the U-Net network model is that: (1) Considering that the SWIM omnidirectional wave height spectrum data has negative spectral values, the SWIM omnidirectional wave height spectrum correction network model uses the leakyReLU activation function in the encoder part to learn the features of negative spectral values ​​for subsequent correction; compared with the use of the ReLU activation function in the encoder part, this model can more effectively correct negative spectral values. (2) The SWIM omnidirectional wave height spectrum correction network model adds a batch normalization (BN) layer after the convolution layer and the transposed convolution layer to solve the problem of internal covariate drift, thereby accelerating the learning speed of the network model and reducing the sensitivity of the network model to the initial learning rate.

[0044] The SWIM omnidirectional wave height spectrum correction network model combines the leakyReLU activation function of the encoder part and the ReLU activation function of the decoder part, so that the negative spectral value of the SWIM omnidirectional wave height spectrum can be effectively corrected; through the jump connection between the encoder and decoder, the extraction and fusion of feature information before and after correction in different dimensions are realized, so that the network model can effectively correct the SWIM omnidirectional wave height spectrum.

[0045] Furthermore, the SWIM omnidirectional wave height spectrum correction network model structure is shown in the figure below: Figure 2As shown in Figure 2, the network input and output are the uncorrected SWIM omnidirectional wave height spectrum and the corrected SWIM omnidirectional wave height spectrum, respectively. The BU-Net network structure is divided into an encoder and a decoder. The encoder includes multiple repeated structures, each of which includes: a convolutional layer, a batch normalization layer (connected after each convolutional layer), an activation function layer, and a pooling layer; the decoder includes multiple repeated structures, each of which includes: a transposed convolutional layer, a convolutional layer, a batch normalization layer (connected after each transposed convolutional layer and convolutional layer), and an activation function layer.

[0046] Preferably, if Figure 2 As shown, the encoder consists of a repetitive structure consisting of two 1×3 convolutional layers, two batch normalization (BN) layers, two activation layers, and a pooling layer. The length and width of the input and output of the convolutional layers remain constant, and the number of convolution kernels determines the number of output channels. A BN layer is typically added after the convolutional layers and before the activation unit to accelerate training. Considering that the input omnidirectional wave height spectrum will have values ​​less than 0, the encoder uses a leaky ReLU activation function to preserve negative features in this region. The convolution kernel size is 1×3 with a stride of 1. Downsampling is achieved through max pooling with a pool size of 1×2 and a stride of 1×2, halving the image size. After each downsampling, the number of convolution kernels is doubled, and the above structure is repeated. The encoding process on the left continuously compresses the width of the features. In the decoder, upsampling is performed through transposed convolution. After each upsampling, the number of convolution kernels is halved. The upsampled result is fused with the corresponding features from the encoder, and then convolution is performed. The size and stride of the convolution kernel are the same as those in the encoder part. The activation function in the decoder part is ReLU.

[0047] In order to quantitatively analyze the correction effect of the omnidirectional wave height spectrum, the evaluation indicators include the correlation coefficient between the SWIM average omnidirectional wave height spectrum and the average omnidirectional wave height spectrum of the corresponding buoy before or after correction, the integrated energy error, and the relative error of the peak wave number.

[0048] Correlation coefficient CC (in Figure 3-4 C) represents the correlation between the SWIM average omnidirectional wave height spectrum and the corresponding NDBC buoy average omnidirectional wave height spectrum before or after correction. The calculation method is as follows:

[0049]

[0050] Where N represents the wave number point, Indicates that at wave number point k i The average omnidirectional wave height spectrum before SWIM correction or the average omnidirectional wave height spectrum after SWIM correction by neural network, F buoy (k i) indicates that at wave number point k i The average omnidirectional wave height spectrum of the buoy at μ A , μ B Represents N wave number points respectively F buoy (k i ) is the mean of .

[0051] The integrated energy relative error ΔE is used to quantitatively describe the energy difference between the SWIM average omnidirectional wave height spectrum and the corresponding NDBC buoy average omnidirectional wave height spectrum before or after correction. The integrated energy error is calculated as follows:

[0052]

[0053] Among them, k i 、 and F buoy (k) has the same meaning as above, Δk i is an equal wavenumber interval in a uniform wavenumber domain.

[0054] Peak wave number relative error Δk p It is used to quantitatively describe the degree of deviation of the peak wavenumber of the SWIM average omnidirectional wave height spectrum relative to the peak wavenumber of the corresponding NDBC buoy average omnidirectional wave height spectrum before or after correction. p Defined as:

[0055]

[0056] Among them, k p_SWIM and k p_buoy are the peak wavenumbers of the SWIM average omnidirectional wave height spectrum and the corresponding NDBC buoy average omnidirectional wave height spectrum before and after correction, respectively. The peak wavenumber of the SWIM average omnidirectional wave height spectrum before correction is the wavenumber at the corresponding spectrum peak after removing spurious peaks, while the peak wavenumber of the SWIM average omnidirectional wave height spectrum after correction is determined directly from the spectrum peak.

[0057] The SWIM omnidirectional wave height spectrum correction network model was trained using the training method provided by this invention, running on an NVIDIA TITAN X discrete graphics card using MATLAB 2022b. The neural network training batch size was set to 256, the Adam optimizer was selected, the initial learning rate was set to 0.01, the mean square error (MSE) loss function was calculated for the regression layer, and L2 regularization was used to prevent overfitting. a1 = 8, a2 = 10; the hyperparameter α of the LeakyReLU activation function in the encoder portion of each beam network model was set to 0.01. α represents the slope of the activation function when the input data is less than 0.

[0058] The SWIM omnidirectional wave height spectrum to be corrected is input into the trained SWIM omnidirectional wave height spectrum correction network model. Under single wind wave conditions and single surge conditions, the comparison results of the average omnidirectional wave height spectra of SWIM and NDBC buoys before and after correction are as follows: Figure 3 、 4 shown.

[0059] Figure 3 Under various sea conditions shown in (a) to (f) of the figure, the minimum correlation coefficient after 6° beam correction is 0.96, which is greater than the correlation coefficient before correction. The absolute value of the integrated energy error ΔE between the average omnidirectional wave height spectrum after SWIM 6° beam correction and the buoy data is less than 20%. The integrated energy error anomaly caused by the pseudo-peak in small sea conditions is corrected. The relative error Δk between the peak wave number after SWIM 6° beam correction and the buoy data is p The absolute values ​​of are all less than 10%, which reduces the degree of peak wave number shift caused by the surfboard effect.

[0060] Figure 4 Under various sea surface conditions shown in Figures (a) to (f), the minimum correlation coefficient after 6° beam correction is 0.97, which is greater than the correlation coefficient before correction. The absolute value of the integrated energy error after SWIM 6° beam correction is less than 10%. From the perspective of the relative error of peak wave number, the average omnidirectional wave height spectrum of SWIM 6° beam after correction is Δk p The absolute values ​​of are all less than 10%. Therefore, under pure swell sea conditions, the SWIM omnidirectional wave height spectrum correction network model has a good correction effect on the false peaks and peak wavenumber shifts in the SWIM omnidirectional wave height spectrum.

[0061] An embodiment of the present invention provides a method for correcting a satellite-borne spectrometer omnidirectional wave height spectrum, which inputs a SWIM omnidirectional wave height spectrum to be corrected into a satellite-borne spectrometer omnidirectional wave height spectrum correction model trained using the method described in any of the above embodiments to obtain a corrected omnidirectional wave height spectrum.

[0062] An embodiment of the present invention provides a training system for an omnidirectional wave height spectrum correction model of a satellite-borne spectrometer, comprising: a computer-readable storage medium and a processor;

[0063] The computer-readable storage medium is used to store executable instructions;

[0064] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the training method as described in any one of the above embodiments.

[0065] An embodiment of the present invention provides an omnidirectional wave height spectrum correction system for a satellite-borne spectrometer, comprising: a computer-readable storage medium and a processor;

[0066] The computer-readable storage medium is used to store executable instructions;

[0067] The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the correction method described in the above embodiment.

[0068] An embodiment of the present invention provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a processor to execute the training method described in any of the above embodiments, or execute the correction method described in the above embodiments.

[0069] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A training method for an omnidirectional wave height spectrum correction model for a satellite-borne spectrometer, characterized in that: include: S1, pre-processing the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy to make them correspond one to one; The preprocessing includes: unifying the wavenumber domain, equalizing and standardizing the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the buoy in sequence; the wavenumber domain is the common wavenumber range of the two; S2, construct a loss function to minimize the difference between the SWIM omnidirectional wave height spectrum to be corrected measured by the satellite-borne spectrometer and the omnidirectional wave height spectrum measured by the corresponding buoy, and train the satellite-borne spectrometer omnidirectional wave height spectrum correction model; The satellite-borne spectrometer omnidirectional wave height spectrum correction model is a neural network model; the loss function is: M is the batch size of training data; N is the number of wavenumber points in the unified wavenumber domain; k j is the jth wavenumber point in the unified wavenumber domain; is the peak wave number of the omnidirectional wave height spectrum measured by the buoy corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; is the SWIM omnidirectional wave height spectrum of the i-th sample after correction by the neural network model; is the omnidirectional wave height spectrum measured by the buoy corresponding to the i-th SWIM omnidirectional wave height spectrum to be corrected in a batch of samples; η * i is the dimensionless variance corresponding to the i-th spectrum to be corrected in a batch of samples; w ij It is a penalty coefficient based on the sea conditions and the relationship between the wave number and the peak wave number. a1 and a2 are the penalty coefficients of the high wave number part of the SWIM omnidirectional wave height spectrum in pure wind waves and pure swell waves, respectively.

2. The method according to claim 1, wherein The neural network model is a BU-Net network model; the BU-Net network model is obtained by processing the U-Net network model; The processing includes: adding a batch normalization layer after each convolutional layer in the encoder, each convolutional layer in the decoder and the transposed convolutional layer of the U-Net network model, and using a leakyReLU activation function as the activation function of the encoder.

3. A method for correcting the omnidirectional wave height spectrum of a satellite-borne spectrometer, characterized in that: The SWIM omnidirectional wave height spectrum to be corrected is input into a satellite-borne spectrometer omnidirectional wave height spectrum correction model trained by the training method according to any one of claims 1 to 2 to obtain a corrected omnidirectional wave height spectrum.

4. A training system for an omnidirectional wave height spectrum correction model of a satellite-borne spectrometer, characterized in that: include: Computer-readable storage media and processor; The computer-readable storage medium is used to store executable instructions; The processor is used to read the executable instructions stored in the computer-readable storage medium and execute the training method according to any one of claims 1-2.

5. A satellite-borne spectrometer omnidirectional wave height spectrum correction system, characterized in that: include: Computer-readable storage media and processor; The computer-readable storage medium is used to store executable instructions; The processor is configured to read the executable instructions stored in the computer-readable storage medium and execute the correction method according to claim 3.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the training method according to any one of claims 1-2, or execute the correction method according to claim 3.

Images