CN117647808A - Non-coherent radar phase analysis wave calendar inversion method based on deep learning - Google Patents
Non-coherent radar phase analysis wave calendar inversion method based on deep learning Download PDFInfo
- Publication number
- CN117647808A CN117647808A CN202410102035.XA CN202410102035A CN117647808A CN 117647808 A CN117647808 A CN 117647808A CN 202410102035 A CN202410102035 A CN 202410102035A CN 117647808 A CN117647808 A CN 117647808A
- Authority
- CN
- China
- Prior art keywords
- radar
- wave
- data
- deep learning
- spectrum
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000013135 deep learning Methods 0.000 title claims abstract description 34
- 238000004458 analytical method Methods 0.000 title claims abstract description 16
- 238000001228 spectrum Methods 0.000 claims abstract description 73
- 238000013507 mapping Methods 0.000 claims abstract description 20
- 230000001427 coherent effect Effects 0.000 claims abstract description 15
- 230000009466 transformation Effects 0.000 claims abstract description 12
- 238000001514 detection method Methods 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims abstract description 4
- 230000017105 transposition Effects 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 34
- 238000001914 filtration Methods 0.000 claims description 12
- 238000004088 simulation Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 7
- 239000006185 dispersion Substances 0.000 claims description 7
- 238000007405 data analysis Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims 1
- 230000003935 attention Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000011160 research Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002592 echocardiography Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000009304 pastoral farming Methods 0.000 description 1
- 230000010332 selective attention Effects 0.000 description 1
- 239000002344 surface layer Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Abstract
The invention discloses a non-coherent radar phase analysis wave calendar inversion method based on deep learning, which relates to the field of ocean remote sensing detection and comprises the following steps: preprocessing radar sea clutter images, and changing gray data of the radar sea clutter images in the region into radar image frequency spectrum data through Fourier transformation; constructing a deep learning spectrum mapping model, and inputting radar image spectrum data into the constructed deep learning spectrum mapping model, wherein the deep learning spectrum mapping model comprises a three-dimensional convolution module, a position coding module, an attention module, a residual module and a three-dimensional transposition convolution module; and finally, carrying out data inversion processing, and converting the frequency spectrum data calculated by the model into actual phase analysis wave calendar data. The invention can reduce the phenomenon of frequency spectrum data missing and improve the accuracy of inversion data.
Description
Technical Field
The invention relates to the field of ocean remote sensing detection, in particular to a radar image sea wave information inversion research based on deep learning.
Background
In the case of grazing incidence, the X-band radar forms a sea clutter image by receiving echoes resonating with the sea surface wave by the radar antenna. The sea clutter image is due to the fact that the wind waves and the swell modulate radar back scattering signals, the radar sea clutter image implies wind waves and swell information, the radar image is rich in sea wave information, the time-space resolution ratio of the radar image is relatively high, sea wave parameters can be extracted from the radar image, and therefore a research hot spot for sensing the ocean environment by using the X-band radar is utilized.
How to extract the implicit sea wave information from the sea clutter image is a very complex process, so many scholars currently conduct a great deal of research on sea wave statistical parameters such as sense wave height, characteristic period, wave direction, sea surface layer flow and the like, and relatively mature sea wave monitoring systems such as a Wamos sea wave detection system in Germany and a WAVEX sea wave detection system in the Netherlands have been established. However, with the development of ocean technology, the measurement of wave statistics is difficult to meet the operation requirement of the current ocean structure, and the measurement of more accurate and finer ocean wave information becomes a key problem to be solved urgently.
The research of phase analysis wave inversion of the non-coherent radar is established and developed abroad, and preliminary results are obtained. It still has the following problems:
in the existing inversion method, the effect that the image spectrum is changed into the actual wave spectrum by Fourier transformation is required to be modulated is mainly that the modulation function is basically a unitary power function, the unitary function cannot completely meet the conversion relation from the image spectrum to the wave spectrum, the phenomenon of frequency spectrum data deletion can occur, and the effect of time calendar data inversion of the wave surface is poor.
Disclosure of Invention
The invention aims to provide a phase analysis wave calendar inversion method based on a non-coherent radar image, which can more accurately invert wave data and reduce errors.
The invention aims to achieve the aim, and the aim is achieved by the following technical scheme:
the non-coherent radar phase analysis wave calendar inversion method based on deep learning comprises the following steps:
preprocessing radar sea clutter images, and changing gray data of the radar sea clutter images in the region into radar image frequency spectrum data through Fourier transformation;
constructing a deep learning spectrum mapping model, and inputting radar image spectrum data into the constructed deep learning spectrum mapping model, wherein the deep learning spectrum mapping model comprises a three-dimensional convolution module, a position coding module, an attention module, a residual error module and a three-dimensional transposition convolution module;
and (3) carrying out data inversion processing, and converting the frequency spectrum data calculated by the model into actual phase analysis wave calendar data.
Preferably, the verification method is added after the inversion method step:
and verifying the accuracy of the deep learning spectrum mapping model through simulation data, open source radar data and real sea experimental measurement data.
The method for constructing the deep learning spectrum mapping model comprises the following steps:
convolving the radar image spectrum data obtained by Fourier transform through Conv 3D;
flattening the input data into a one-dimensional vector after three-dimensional convolution, and adding position codes into the one-dimensional vector;
adding the position-coded one-dimensional vector into an attention module, and giving weight;
adding a residual error module before and after the attention module, specifically expressed as:
wherein,is depth +.>Is (are) unit features of->Is depth +.>Is (are) unit features of->Is->Input vector->For transfer to->Weights of the individual input vectors, +.>Is an arbitrary function;
the size is recovered using three-dimensional transpose convolution.
The step of converting the radar image gray data into radar image spectrum data includes:
selecting a data analysis area according to the input radar sea clutter image dataWherein->For the initial radar sea clutter image, +.>For selecting radar sea clutter images of the region, r is the distance between the measuring point and the origin of coordinates, +.>The azimuth angle of the antenna is, t is radar measurement time, and R is radar detection distance;
according to the radar image in the selected area, completing time-space domain-frequency domain conversion through Fourier transformation, and carrying out three-dimensional Fourier transformation on the processed radar sea clutter image sequence to obtain a wave number frequency spectrum:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For wave number in the horizontal direction of space>For wave number in the vertical direction of space, +.>For the transformed frequency, +.>For the rectangular coordinate expression of radar sea clutter image, < >>，，/>The inversion space interval of the selected region is T which is the time length;
and filtering the radar image spectrum data to eliminate the influence of other noise and improve the inversion calculation precision.
Since the wave frequency distribution range of the wave frequency characteristics is above 0.3, the starting frequency of the high pass filter is 0.3, which can be expressed as:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For the converted frequency
Because the wave frequency satisfies the dispersion relation, the frequency is distributed around the dispersion relation belt, and the dispersion relation is expressed asSo a band pass filter bandwidth of 0.2 is expressed as:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For union sign +.>Is the wave number in the horizontal direction in space,for wave number in the vertical direction of space, +.>G is gravitational acceleration, which is the converted frequency.
The spectrum data is converted into a 5-order tensor form, input into a five-dimensional tensor (b×t×v×h×w), and output into a five-dimensional tensor (b×t×d×h×w), where B is a lot, V is a variable, D is a depth, H is a longitude, and W is a latitude.
Conv3D convolution is specifically expressed as:
wherein,V、H、Wfor the length of the data in the horizontal, height and depth directions,v、h、wis thatV、H、WIs used for the indexing of (a),Bin order to input a sample of the sample,Tin the form of a convolution kernel,for the jth convolution kernelv、h、wThe weight value of the time-out is calculated,for the ith samplev、h、wData values, ->The bias term for the jth convolution kernel,is the convolution result of the ith sample under the jth convolution kernel.
The step of recovering the size using three-dimensional transpose convolution includes:
performing inverse Fourier transform on the sea wave spectrum calculated and output by the model to finish conversion inversion from a frequency domain to a time domain to obtain a wave surface calendar, wherein the inverse transform is as follows:
wherein the method comprises the steps ofCalculating the output ocean wave spectrum by the model>I is a complex index for the final wave surface epoch.
The specific expression of the position code is:
wherein,for the position index of the input vector, < >>For the dimension index in the position-coding vector, +.>For feature vector dimension, < >>Position coding for even positions, +.>Position codes for odd positions.
The step of verifying accuracy includes:
repeating the steps of the inversion method on different input data, and judging the final result through root mean square error and correlation coefficient, wherein the root mean square error mse is as follows:
the correlation coefficient corr is:
wherein,representing the wave surface calculated by inversion +.>Representing the actual wave surface, n representing the number of sampling points.
The invention has the advantages that:
the invention provides a novel non-coherent radar phase analysis wave inversion method, which constructs the mapping relation between radar spectrum and sea wave spectrum through special deep learning, and the problem can be overcome by the patent, and the high-precision phase analysis wave inversion can be realized because the phenomenon that the data is true when the mapping transformation between the two spectrums is realized only by adopting an empirical function in the existing method is influenced by the phenomenon that the data is true
The invention adopts three-dimensional convolution in sea wave inversion, and for two-dimensional convolution, the same filter is output as a two-dimensional characteristic diagram, and the information of multiple channels is completely compressed, so that the information on a time sequence cannot be well captured. The output of the three-dimensional convolution is still a 3D feature map, which can better capture the data space and time features.
In addition, as the input data are flattened into one-dimensional vectors after three-dimensional convolution, in order to prevent the position information from being lost after the data are flattened, the scheme is further added with position codes after the data are flattened and then transmitted into the attention module, so that the mapping between data matrixes is realized, the phenomenon of frequency spectrum data loss is reduced, and the accuracy of inversion data is improved.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of a deep learning model structure according to the present invention;
FIG. 3 is a two-dimensional effect diagram of simulated radar data for a position-based sea condition in accordance with an embodiment of the present invention;
FIG. 4 is a two-dimensional effect diagram of position two-simulation radar data under a four-level sea condition according to an embodiment of the present invention;
FIG. 5 is a two-dimensional effect diagram of three-simulation radar data of a position under a four-level sea condition according to an embodiment of the present invention;
fig. 6 is a two-dimensional effect diagram of four-simulation radar data of a position under four-level sea conditions according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
The embodiment discloses a non-coherent radar phase analysis wave calendar inversion method based on deep learning, please refer to the flow of fig. 1, which specifically comprises the following steps.
S1, preprocessing radar sea clutter images, and changing gray data of the radar sea clutter images in the area into radar image frequency spectrum data through Fourier transformation.
(1) According to the input radar sea clutter image dataSelecting a suitable radar data analysis area, wherein +.>For the initial radar sea clutter image, +.>For selecting radar sea clutter images of the region, r is the distance between the measuring point and the origin of coordinates, +.>For the azimuth angle of the antenna, t is radar measurement time, R is radar detection distance, the radial distance of the selected area of the invention is (800 m-1800 m), the spatial distance resolution is 5m, the direction angle is (20-60 degrees), 32 sea clutter images are obtained in total, each image is separated by 2s, and the total time is 64s.
(2) According to the radar image in the selected area, completing time-space domain-frequency domain conversion through Fourier transformation, and carrying out three-dimensional Fourier transformation on the processed radar sea clutter image sequence to obtain a wave number frequency spectrum:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For wave number in the horizontal direction of space>For wave number in the vertical direction of space, +.>For the transformed frequency, +.>For radar sea clutter imagesRectangular coordinate expression of>，，/>The space interval is the inversion space interval of the selected region, and T is the time length.
S2, carrying out band-pass filtering and high-pass filtering on the obtained radar spectrum data, filtering redundant noise in the data, and inputting the data into a constructed deep learning model
(1) And filtering the obtained radar spectrum by adopting a high-pass filter, and filtering noise data, wherein the filtering operation is as follows:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For the converted frequency
(2) The radar spectrum is filtered by adopting the band-pass filtering function, and the frequency of the radar spectrum is distributed around the dispersion relation belt due to the fact that the wave frequency meets the dispersion relation, and the dispersion relation is expressed asSo a band pass filter bandwidth of 0.2 is expressed as:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For union sign +.>Is the wave number in the horizontal direction in space,for wave number in the vertical direction of space, +.>G is gravitational acceleration, which is the converted frequency.
(3) The spectrum data is converted into a 5-order tensor form, input into a five-dimensional tensor (b×t×v×h×w), output into a five-dimensional tensor (b×t×d×h×w), B is a lot, V is a variable, D is a depth, H is a longitude, and W is a latitude.
S3, constructing a deep learning spectrum mapping model, referring to FIG. 2, inputting radar image spectrum data into the constructed deep learning spectrum mapping model, wherein the deep learning spectrum mapping model comprises a three-dimensional convolution module, a position coding, an attention module, a residual error module and a three-dimensional transposition convolution module.
(1) The radar image spectrum data obtained by Fourier transform is convolved through Conv3D, the data is matched with the input and output of the model according to the radar image spectrum data obtained by Fourier transform, the input spectrum data is convolved through Conv3D, and the 3D convolution is changed on the basis of 2D convolution. For 2D convolution, the same filter output is a two-dimensional feature map, and the information of multiple channels is completely compressed, so that the information on the time sequence cannot be captured well. The output of the 3D convolution is still a 3D feature map, which enables better capture of data spatial and temporal features.
The input multidimensional sea surface environment data is convolved by a 3D convolution module, and the output of the 3D convolution is still a 3D characteristic diagram; the size of the convolution kernel in the 3D convolution is (2, 5, 5), the step size is (1, 1, 1), and the concrete expression of Conv3D convolution is:
wherein,V、H、Wfor the length of the data in the horizontal, height and depth directions,v、h、wis thatV、H、WIs used for the indexing of (a),Bin order to input a sample of the sample,Tin the form of a convolution kernel,for the jth convolution kernelv、h、wThe weight value of the time-out is calculated,for the ith samplev、h、wData values, ->The bias term for the jth convolution kernel,is the convolution result of the ith sample under the jth convolution kernel.
(2) Calculating position codes, flattening input data into one-dimensional vectors after three-dimensional convolution, and adding the position codes into the one-dimensional vectors in order to prevent position information from being lost after the data are flattened:
wherein,for the position index of the input vector, < >>For the dimension index in the position-coding vector, +.>For feature vector dimension, < >>Is an even number positionPosition coding of->Position codes for odd positions.
(3) The one-dimensional vector after the position coding is added into an attention module, weight is given, the attention module simulates the selective attention of human beings, more key information is gradually selected in the training process, different weights are given according to the importance degree of different information, and the specific expression of the attention is as follows:
wherein,for the i-th input vector,/->For the ith query vector, < >>For the ith key vector, < >>For the i-th value vector,/->、/>、/>Is a trainable weight matrix, +.>For square root of dimension of M, Q is the input query vector, M is the input key vector, V is the input value directionQuantity, T denotes transpose, ">In order to be able to take care of the size of the attention,the expression of (2) is:
wherein,representing the similarity of the ith key vector and the query vector,/->Representing the similarity of the jth key vector and the query vector, n is the total number of data, exp represents the natural exponential function.
(4) Adding a residual error module before and after the attention module, specifically expressed as:
wherein,is depth +.>Is (are) unit features of->Is depth +.>Is (are) unit features of->Is->Input vector->For transfer to->Weights of the individual input vectors, +.>Is an arbitrary function;
(5) The size is recovered by using three-dimensional transposed convolution, the size of a convolution kernel in the 3D transposed convolution is (2, 5, 5), the step length is (2, 1, 1), and the transposed convolution is specifically expressed as:
wherein I is an output vector, O is an input vector, and C is a convolution kernel matrix.
S4, data inversion processing is carried out, and spectrum data calculated by the model are converted into actual phase analysis wave calendar data.
Performing inverse Fourier transform on the sea wave spectrum calculated and output by the model to finish conversion inversion from a frequency domain to a time domain to obtain a wave surface calendar, wherein the inverse transform is as follows:
wherein the method comprises the steps ofCalculating the output ocean wave spectrum by the model>I is a complex index for the final wave surface epoch.
S5, verifying the accuracy of the deep learning spectrum mapping model through simulation data and real sea experimental measurement data.
Repeating the steps of the inversion method on different input data, and judging the final result through root mean square error and correlation coefficient, wherein the root mean square error mse is as follows:
the correlation coefficient corr is:
wherein,representing the wave surface calculated by inversion +.>The actual wave surface is represented, n represents the number of sampling points, and the embodiment requires mse not higher than 15% and corr not lower than 0.85 to meet the requirement.
Fig. 3, 4, 5, 6 and table 1 show phase resolved wave results calculated from numerical simulation radar image data.
Table 1 error calculation table
The simulation adopts a general simulation method, firstly, wave fields under four-level sea conditions (sense wave height is 1.5 m) and five-level sea conditions (sense wave height is 2.5 m) are obtained through simulation, and then, a shadow modulation tilt modulation and a bragg scattering mechanism are introduced to obtain radar images under the four-level sea conditions.
According to the result, the calculated correlation coefficient is more than 0.85, and the relative error is less than 15%, so that the effectiveness of the method can be proved.
Finally, it should be noted that: the foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (9)
1. The non-coherent radar phase analysis wave calendar inversion method based on the deep learning is characterized in that the actual physical problem is solved by an intelligent method of the deep learning to realize high-precision phase analysis wave inversion, and the method comprises the following steps:
preprocessing radar sea clutter images, selecting an analysis area, and changing gray data of the radar sea clutter images in the area into radar image frequency spectrum data through Fourier transformation;
performing band-pass filtering and high-pass filtering operation on the obtained radar spectrum data, filtering redundant noise in the radar spectrum data, and retaining main sea wave information;
constructing a deep learning spectrum mapping model, and inputting radar image spectrum data into the constructed deep learning spectrum mapping model, wherein the deep learning spectrum mapping model comprises a three-dimensional convolution module, a position coding module, an attention module, a residual error module and a three-dimensional transposition convolution module;
and (3) carrying out data inverse processing, namely converting the frequency spectrum data calculated by the model into actual phase analysis wave calendar data through inverse Fourier transform.
2. The deep learning based non-coherent radar phase resolved wave calendar inversion method of claim 1, further comprising the steps of:
and verifying the accuracy of the deep learning spectrum mapping model through simulation data and real sea experimental measurement data.
3. The deep learning-based non-coherent radar phase-resolved wave calendar inversion method according to claim 1, wherein the step of constructing a deep learning spectrum mapping model is:
convolving the radar image spectrum data obtained by Fourier transform through Conv 3D;
flattening the input data into a one-dimensional vector after three-dimensional convolution, and adding position codes into the one-dimensional vector;
adding the position-coded one-dimensional vector into an attention module, and giving weight;
adding a residual error module before and after the attention module, specifically expressed as:
wherein,is depth +.>Is (are) unit features of->Is depth +.>Is (are) unit features of->Is->Input vector->For transfer to->Weights of the individual input vectors, +.>Is an arbitrary function;
the size is recovered using three-dimensional transpose convolution.
4. The deep learning based non-coherent radar phase resolved wave calendar inversion method of claim 1, wherein the step of changing radar image gray data into radar image spectral data comprises:
selecting a data analysis area according to the input radar sea clutter image dataWhereinFor the initial radar sea clutter image, +.>For selecting radar sea clutter images of the region, r is the distance between the measuring point and the origin of coordinates, +.>The azimuth angle of the antenna is, t is radar measurement time, and R is radar detection distance;
according to the radar image in the selected area, completing time-space domain-frequency domain conversion through Fourier transformation, and carrying out three-dimensional Fourier transformation on the processed radar sea clutter image sequence to obtain a wave number frequency spectrum:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For wave number in the horizontal direction of space>For wave number in the vertical direction of space, +.>For the transformed frequency, +.>For the rectangular coordinate expression of radar sea clutter image, < >>，，/>The space interval is the inversion space interval of the selected region, and T is the time length.
5. The deep learning based non-coherent radar phase resolved wave calendar inversion method of claim 4, wherein the step of filtering the radar image spectral data comprises:
wave frequency characteristics the wave frequency distribution range is above 0.3, the high pass filter initial frequency is 0.3, expressed as:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>Is the converted frequency;
the dispersion relation is expressed asThe band pass filter bandwidth is 0.2, expressed as:
wherein the method comprises the steps ofFor the transformed wave number frequency spectrum, +.>For wave number in the horizontal direction of space>For wave number in the vertical direction of space, +.>Is the converted frequency;
the spectrum data is converted into a 5-order tensor form, input into a five-dimensional tensor (b×t×v×h×w), and output into a five-dimensional tensor (b×t×d×h×w), wherein B is a lot, V is a variable, D is depth, H is a horizontal direction length, and W is a water direction length.
6. The deep learning-based non-coherent radar phase-resolved wave calendar inversion method according to claim 3, wherein Conv3D convolution is specifically expressed as:
wherein,V、H、Wfor the length of the data in the horizontal, height and depth directions,v、h、wis thatV、H、WIs used for the indexing of (a),Bin order to input a sample of the sample,Tin the form of a convolution kernel,for the jth convolution kernelv、h、wWeight value of time,/->For the ith samplev、h、wData values, ->Bias term for jth convolution kernel, < +.>For the ith sample at the ithConvolution results under j convolution kernels.
7. The deep learning based non-coherent radar phase resolved wave calendar inversion method of claim 3, wherein said step of recovering dimensions using three-dimensional transpose convolution comprises:
performing inverse Fourier transform on the sea wave spectrum calculated and output by the model to finish conversion inversion from a frequency domain to a time domain to obtain a wave surface calendar, wherein the inverse transform is as follows:
wherein the method comprises the steps ofCalculating the output ocean wave spectrum by the model>I is a complex index for the final wave surface epoch.
8. The deep learning-based non-coherent radar phase resolved wave calendar inversion method according to claim 3, wherein the specific expression of the position code is:
wherein,for the position index of the input vector, < >>Encoding vectors for positionsDimension index of>For feature vector dimension, < >>Position coding for even positions, +.>Position codes for odd positions.
9. The deep learning based non-coherent radar phase resolved wave calendar inversion method of claim 2, wherein the step of verifying accuracy comprises:
repeating the steps of the inversion method on different input data, and judging the final result through root mean square error and correlation coefficient, wherein the root mean square error mse is as follows:
the correlation coefficient corr is:
wherein,representing the wave surface calculated by inversion +.>Representing the actual wave surface, n representing the number of sampling points.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410102035.XA CN117647808B (en) | 2024-01-25 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410102035.XA CN117647808B (en) | 2024-01-25 | Non-coherent radar phase analysis wave calendar inversion method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117647808A true CN117647808A (en) | 2024-03-05 |
| CN117647808B CN117647808B (en) | 2024-04-30 |
Family
ID=
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103969643A (en) * | 2014-05-09 | 2014-08-06 | 哈尔滨工程大学 | Method for conducting X wave band navigation radar wave parameter inversion through band-pass filter based on novel wave dispersion relation |
| CN104062635A (en) * | 2014-07-04 | 2014-09-24 | 哈尔滨工程大学 | Sea clutter image sea wave principal wave period retrieval method based on minimum interval thought |
| US20170315225A1 (en) * | 2016-04-28 | 2017-11-02 | Electronics And Telecommunications Research Institute | Apparatus and method for extracting ocean wave information |
| CN109557538A (en) * | 2018-12-26 | 2019-04-02 | 哈尔滨工业大学 | The method for measuring ocean wave parameter with coherent radar based on sea |
| US20220374681A1 (en) * | 2022-07-08 | 2022-11-24 | Zhihan Lv | Method for predicting wave energy based on improved GRU |
| CN115755043A (en) * | 2022-10-19 | 2023-03-07 | 华中科技大学 | Wave field reconstruction and prediction method based on X-band non-coherent radar |
| CN116202491A (en) * | 2023-02-24 | 2023-06-02 | 青岛海研电子有限公司 | Ocean wave direction spectrum drifting wave buoy and ocean wave direction spectrum estimation method |
| CN116306318A (en) * | 2023-05-12 | 2023-06-23 | 青岛哈尔滨工程大学创新发展中心 | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning |
| CN116484271A (en) * | 2023-03-20 | 2023-07-25 | 上海大学 | Effective wave height early warning method based on empirical mode decomposition and deep learning |
| CN116486294A (en) * | 2022-12-07 | 2023-07-25 | 大连海事大学 | Wave level identification method integrating convolution attention mechanism and InceptionResNet |
| CN116500611A (en) * | 2023-03-21 | 2023-07-28 | 大连理工大学 | Deep learning-based radar wave surface image sea wave parameter inversion method |
| CN116933152A (en) * | 2023-06-07 | 2023-10-24 | 哈尔滨工业大学（威海） | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network |
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103969643A (en) * | 2014-05-09 | 2014-08-06 | 哈尔滨工程大学 | Method for conducting X wave band navigation radar wave parameter inversion through band-pass filter based on novel wave dispersion relation |
| CN104062635A (en) * | 2014-07-04 | 2014-09-24 | 哈尔滨工程大学 | Sea clutter image sea wave principal wave period retrieval method based on minimum interval thought |
| US20170315225A1 (en) * | 2016-04-28 | 2017-11-02 | Electronics And Telecommunications Research Institute | Apparatus and method for extracting ocean wave information |
| CN109557538A (en) * | 2018-12-26 | 2019-04-02 | 哈尔滨工业大学 | The method for measuring ocean wave parameter with coherent radar based on sea |
| US20220374681A1 (en) * | 2022-07-08 | 2022-11-24 | Zhihan Lv | Method for predicting wave energy based on improved GRU |
| CN115755043A (en) * | 2022-10-19 | 2023-03-07 | 华中科技大学 | Wave field reconstruction and prediction method based on X-band non-coherent radar |
| CN116486294A (en) * | 2022-12-07 | 2023-07-25 | 大连海事大学 | Wave level identification method integrating convolution attention mechanism and InceptionResNet |
| CN116202491A (en) * | 2023-02-24 | 2023-06-02 | 青岛海研电子有限公司 | Ocean wave direction spectrum drifting wave buoy and ocean wave direction spectrum estimation method |
| CN116484271A (en) * | 2023-03-20 | 2023-07-25 | 上海大学 | Effective wave height early warning method based on empirical mode decomposition and deep learning |
| CN116500611A (en) * | 2023-03-21 | 2023-07-28 | 大连理工大学 | Deep learning-based radar wave surface image sea wave parameter inversion method |
| CN116306318A (en) * | 2023-05-12 | 2023-06-23 | 青岛哈尔滨工程大学创新发展中心 | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning |
| CN116933152A (en) * | 2023-06-07 | 2023-10-24 | 哈尔滨工业大学（威海） | Wave information prediction method and system based on multidimensional EMD-PSO-LSTM neural network |
Non-Patent Citations (5)
| Title |
|---|
| FANG GAO: "Recognition of ECG Signals by Convolutional Neural Network Based on Attentional Mechanism", 2021 4TH INTERNATIONAL CONFERENCE ON ADVANCED ELECTRONIC MATERIALS, COMPUTERS AND SOFTWARE ENGINEERING (AEMCSE), 19 August 2021 (2021-08-19) * |
| 孙志铭: "基于深度学习的海浪有效波高预报方法研究", 中国硕士学位论文全文数据库 信息科技辑, 15 February 2023 (2023-02-15) * |
| 徐凌宇;张高唯;江湾湾;李新绢;王蕾;: "深度学习神经网络及其在海洋环境信息挖掘预测中的应用", 海洋信息, no. 01, 9 February 2018 (2018-02-09) * |
| 王伟;张斌;马跃华;朱昨庆;: "基于MIMO雷达的海面场景成像模拟及海浪信息反演方法", 中国科学:信息科学, no. 06, 20 June 2016 (2016-06-20) * |
| 王静;唐军武;何宜军;王鑫;潘玉方;: "X-波段雷达近海海浪频谱反演的神经网络模型", 海洋学报(中文版), no. 02, 15 March 2013 (2013-03-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | SwinSUNet: Pure transformer network for remote sensing image change detection | |
| CN106355151B (en) | A kind of three-dimensional S AR images steganalysis method based on depth confidence network | |
| CN104950305B (en) | A kind of real beam scanning radar angle super-resolution imaging method based on sparse constraint | |
| CN102622653B (en) | Multi-resolution path planning method for micro unmanned aerial vehicle under influence of wind field | |
| CN106405548A (en) | Inverse synthetic aperture radar imaging method based on multi-task Bayesian compression perception | |
| CN103984966A (en) | SAR image target recognition method based on sparse representation | |
| CN105954750B (en) | The non-sparse scene imaging method of stripmap synthetic aperture radar based on compressed sensing | |
| Miao et al. | An improved lightweight RetinaNet for ship detection in SAR images | |
| CN106526529A (en) | Sparse representation-based direction-of-arrival estimation method in mismatched condition of steering vectors | |
| CN112257741B (en) | Method for detecting generative anti-false picture based on complex neural network | |
| CN113361367A (en) | Underground target electromagnetic inversion method and system based on deep learning | |
| Liang et al. | Two-dimensional DOA estimation method of acoustic vector sensor array based on sparse recovery | |
| CN109143151A (en) | The uniform surface battle array tensor reconstructing method and signal source locating method of part array element damage | |
| CN112285709A (en) | Atmospheric ozone remote sensing laser radar data fusion method based on deep learning | |
| CN117647808B (en) | Non-coherent radar phase analysis wave calendar inversion method based on deep learning | |
| CN117647808A (en) | Non-coherent radar phase analysis wave calendar inversion method based on deep learning | |
| CN116306318B (en) | Three-dimensional ocean thermal salt field forecasting method, system and equipment based on deep learning | |
| Li et al. | A radar signal recognition approach via IIF-net deep learning models | |
| Yang et al. | GAN-based radar spectrogram augmentation via diversity injection strategy | |
| Zhang et al. | A Gaussian process regression-based sea surface temperature interpolation algorithm | |
| CN102298768B (en) | High-resolution image reconstruction method based on sparse samples | |
| CN105931184A (en) | SAR image super-resolution method based on combined optimization | |
| CN103954960A (en) | Airborne early warning radar moving target detecting method based on clutter ridge guiding vector | |
| Yu et al. | Sl-bilstm: A signal-based bidirectional LSTM network for over-the-horizon target localization | |
| Herchig et al. | Comparing the performance of convolutional neural networks trained to localize underwater sound sources |
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 |