WO2022016884A1

WO2022016884A1 - Method for extracting sea surface wind speed on basis of k-means clustering algorithm

Info

Publication number: WO2022016884A1
Application number: PCT/CN2021/080051
Authority: WO
Inventors: 王慧; 邱海洋; 智鹏飞; 朱琬璐; 朱志宇
Original assignee: 江苏科技大学
Priority date: 2020-07-22
Filing date: 2021-03-10
Publication date: 2022-01-27
Also published as: CN111830506A; CN111830506B

Abstract

Disclosed is a method for extracting a sea surface wind speed from a navigation radar image, wherein the method is based on a K-means clustering algorithm and belongs to the field of using a remote sensing means to perform inversion on a sea surface wind speed. The present invention comprises four parts, i.e. navigation radar image data preprocessing, radar data classification based on a K-means clustering algorithm, sea surface wind speed extraction model determination and sea surface wind speed information extraction. Heterogeneous data is obtained by means of a sea surface wind speed inversion process, and the influence of interference data on a sea surface model is removed, thereby improving the robustness of the model; and for rejected heterogeneous data, a nonlinear quadratic function is used to determine a sea surface wind speed extraction model, thereby improving the precision and speed of sea surface wind speed extraction by the model. Actual measured data is used for verifying the present invention, and in the present invention, a correlation coefficient of a sea surface wind speed and a reference wind speed reaches 0.99, a standard deviation is 0.38 m/s, and a deviation is -0.04 m/s, which are enough to meet engineering and environment monitoring requirements.

Description

A method of sea surface wind speed based on K-means clustering algorithm

technical field

The invention relates to the technical field of sea surface wind speed remote sensing using X-band nautical radar images for sea surface wind speed inversion calculation, in particular to a sea surface wind speed method based on K-means clustering algorithm.

Background technique

The sea surface wind field is an important factor in the study of ocean dynamics and an important guarantee for the safety of navigation operations. It plays a vital role in understanding ocean changes and predicting sea surface risks. The sea surface wind field information mainly includes the sea surface wind direction and the sea surface wind speed. The present invention is a method for extracting the sea surface wind speed information based on the nautical radar image.

The traditional way of extracting wind speed information on the sea surface is mainly anemometers, which are installed on ships, shores or buoys to measure the wind speed. At the same time, this method is susceptible to the influence of sea weather or sea traffic, and lacks continuity in time and space.

Existing remote sensing wind measurement methods mainly include scatterometers, airborne or spaceborne synthetic aperture radar (SAR), satellite altimeters and marine radars, etc. However, scatterometers have the problem of low resolution, low repeated sampling rate of satellite remote sensing, and are subject to The problem of cloud interference means that the measurement data may not be the wind speed information to be detected at the sea surface. X-band marine radar has the advantages of being unaffected by light, capable of real-time continuous feedback and high resolution, and has become one of the important means of marine environmental monitoring at this stage. At present, the X-band marine radar at home and abroad has realized the monitoring of sea surface waves, currents, and rainfall, and the measurement of the oil spill area on the sea surface, but the sea surface wind field measurement based on marine radar images is still in the primary research stage.

At present, there are two main algorithms for inversion of sea surface wind speed based on marine radar images: one is the neural network method, and the other is the model function method. In 2002, Dankert first proposed the neural network method. According to the relationship between the radar scattering cross-sectional area and wind speed, the sea surface wind direction information and NRCS were used as inputs, and the BP neural network was used to invert the sea surface wind speed. In 2006, Dankert considered marine factors such as humidity, temperature, and signal-to-noise ratio as the input of BP neural network to improve the applicability of sea surface wind speed. Jia Ruicai of Harbin Engineering University used the double-hidden-layer unipolar S-function BP neural network method to invert the sea surface wind speed information, which improved the convergence speed of the neural network and the network extension ability. However, the application of neural network has inherent shortcomings. The main problem is the poor applicability of the model. For different environmental locations and different types of marine radars, a large amount of data is required for retraining, and marine environmental factors have a great impact on the method, so the accuracy cannot be achieved. ensure.

In 2005, Horstmann first proposed the application of geophysical model function (GMF), when inputting the SAR echo intensity and sea surface wind direction information, the sea surface wind speed information can be obtained. Although it cannot be directly applied to marine radar, it has been proved that the sea surface wind field and radar echo The intensity has a certain exponential model function relationship. In 2007, Dankert used the measured data to verify the exponential function model of radar echo intensity and sea surface wind speed, but the inversion accuracy could not meet the engineering requirements. In 2012, Lund et al. aimed at the FurunoFAR2117BB marine radar and found that there is a cubic polynomial nonlinear relationship between the RCS and the sea surface wind speed, and used the calculated wind speed to greatly improve the inversion accuracy. In 2013, for the Furuno2117BB radar, Bueno et al. used the linear integration method to obtain the relationship between the radar echo intensity level and the sea surface wind speed function to obtain the sea surface wind speed information. In 2015, Liu Y et al. proposed to use hyperbolic fitting to extract sea surface wind speed information from the measured marine radar data for two radars, Decca and Furono. In 2017, Huang W et al. proposed the use of RCS spectral analysis algorithm, RCS and sea surface wind speed empirical mode decomposition method for Decca radar, and established a function model to obtain sea surface wind speed. In 2015, Chen Zhongbiao and others fit the RCS, effective wave height and sea surface wind speed into a linear probability distribution function for the 9.3GHz Furuno radar, thereby obtaining the sea surface wind speed information. The above methods do not consider the influence of rainfall and marine environmental factors. When the radar image is affected by rainfall, the inversion accuracy and data applicability of the model cannot be guaranteed. The sea surface wind speed inversion model functions generally have problems of high extraction accuracy and poor applicability of sea conditions, which restrict the development prospect of this method.

In view of the above problems, the present invention discloses a method for retrieving sea surface wind speed from marine radar images based on K-means clustering algorithm. First, the radar image is recognized by rainfall image to remove the influence of rainfall on the extraction of sea surface wind speed; secondly, combined with sensors and radar image information to classify the sea surface influencing factor data, eliminate the influence of heterogeneous data on the sea surface wind speed extraction model, and improve the robustness of the model; finally, inheriting the nonlinear relationship between sea surface wind speed and echo intensity proposed by Lund, the Eliminate heterogeneous data and apply nonlinear quadratic function to determine the sea surface wind speed extraction model, which ensures the accuracy of model extraction of sea surface wind speed. The measured data proves the engineering feasibility of the method to extract the sea surface wind speed information from the nautical radar image.

SUMMARY OF THE INVENTION

The invention discloses a sea surface wind speed method based on the K-means clustering algorithm. The method is based on the K-means clustering algorithm, and specifically includes the following steps:

Step 1, radar image data preprocessing. The marine radar monitoring system is used to collect the sea surface radar image sequence data, and the wind meter is used to collect the synchronized sea surface wind direction and wind speed information. Apply Zero Intensity Percentage (ZPP) to radar image sequences to identify and eliminate image data with large rainfall noise; for images with less rain and snow interference, image median filtering is applied to suppress the interference of noise and same-frequency signals on the extraction of sea surface wind direction .

Step 2, data classification based on K-means clustering algorithm. First, the data normalization is performed on the radar image echo intensity, sea surface wind direction information, sea surface wind speed information and the calculated image signal-to-noise ratio, so that the data are in the same coordinate range; secondly, the K-means clustering algorithm is applied to the radar image The image echo intensity, sea surface wind direction information and image signal-to-noise ratio data are classified according to the Euclidean distance, and the error of the centroid distance is used as the judgment basis to obtain heterogeneous data; finally, the radar data and sea surface wind field information data are all excluded. Corresponding information data, the cluster data of radar data and sea surface wind field information are obtained.

Step 3, the sea surface wind speed extraction model is determined. Using clustered radar data and sea surface wind speed data, nonlinear quadratic fitting of sea surface wind speed was performed to obtain the extraction model of sea surface wind speed, and SSE was used to verify the accuracy of the model.

Step 4, extraction of sea surface wind speed information. Select some images of the test marine radar image, carry out normalized mapping, and input them into the sea surface wind speed extraction model to obtain the sea surface wind speed information.

The

steps

2 and 3 of the sea surface wind speed extraction method based on the K-means clustering algorithm include the following steps:

Step 2.1, normalized data processing of the average value of radar echo intensity, sea surface wind direction, wind speed information and image signal-to-noise ratio information;

①Select an appropriate part of the radar image for the preprocessed marine radar image, and perform normalized mapping along the x and y axes to obtain the average value of the radar image f' _i :

Among them, f(x, y) is the selected radar image intensity value, N _x and N _y are the number of pixels along x and y of the selected image, and i is the corresponding radar image number. Normalize f' _i to get the radar image normalized value F _i :

② Obtain the signal-to-noise ratio r _{t of the} selected radar image, normalize it with the radar image time series, and obtain the normalized value R _{i of the} sea state information:

in

is the corrected ocean wave spectrum of the two-dimensional wavenumber spectrum,

is the noise spectrum other than the wave signal in the radar image.

③ Normalize the sea surface wind direction information d _i and the sea surface wind speed information s _i collected by the anemometer according to the radar image sequence, and obtain the normalized values D _i and S _{i of the} sea surface wind direction and wind speed information:

Step 2.2, radar data classification based on K-means clustering algorithm;

① Initialize K initial cluster centroids;

Obtained in step 2.1 F _{_{_i,}} R _i, D _i and S _i of all the data is divided into two sections, one for determining the wind speed based clustering algorithm sea Model K-means, one for the data of the test model; and _{The data F i} , R _i , and D _i used in the determination of the model form a data set, which is used as a set of factors affecting the sea surface wind speed;

Ω _i ={F _i ,R _i ,D _i } (5)

The cluster centroids are initialized, and K data points in the _{Ω i region are randomly selected as the initial centroids.}

② Divide the data points according to the initialized centroid;

After determining the centroids of the K sea surface wind speed influencing factors, find the data points closest to the centroids _{in the data set Ω i, thereby forming clusters.} Here, the Euclidean distance is used for measurement, and _{the data points X i} (x ₁ ,x ₂ ,x ₃ ) of the characteristics of all the factors affecting the sea surface wind speed in _{Ω i} are calculated and the selected K centroids C _k (c ₁ ,c ₂ ,c ₃ ), the formula is as follows:

After each point finds the closest centroid, it belongs to the cluster, and the data set Ω _i is divided into K sub-region spaces Τ _k .

③ Update the cluster centroids;

for each Τ _k

The mean value is used as the centroid of the next update, and the calculation formula is as follows:

According to the updated centroid, the Euclidean distance between the data point and the centroid is recalculated according to formula (6), and a new cluster is formed at the same time.

④The judgment basis for the centroid to stop updating;

According to the distance between the original centroid C _k and the updated centroid C _j , it is determined whether the centroid needs to be further updated. The judgment conditions are as follows:

||C _k -C _j ||＜＜γ (8)

Among them, γ=0.1, when the above conditions are met, it means that the centroids tend to converge, and the classification algorithm is terminated; if the above conditions are not met, steps 2.3 to 2.5 are repeated continuously until formula (8) is satisfied, and the cluster centroids C _f ( f = 1,2, ... f), and each cluster centroid corresponding data set Τ _f.

Step 2.3 remove radar data of heterogeneous data;

According to the obtained cluster distribution of the influencing factors of sea surface wind speed, the centroid whose centroid position is farthest from other centroids is determined as the heterogeneous centroid C _d , and all the data points in its area are also determined as the heterogeneous data set Τ _d , and the data is removed. heterogeneous data sets in Ω _i, S _i the simultaneous removal of heterogeneous data corresponding to a position of surface wind speed S _d, to give the final removal of heterogeneous radar data _{_{_{F f, R f, D f}}} , S f:

Ω _f = {F _f , R _f , D _f } = {Ω _i -Τ _d } S _f ={S _i -S _d } (9)

Step 3.1, determine the sea surface wind speed extraction model; first release the _{original characteristics of F f} , S _f data to obtain the corresponding radar image echo intensity average f _f , and the training sea surface wind speed information s _f :

_{_{f f = F f * max (}} f i '), s f = S f * max (s i) (10)

Step 3.2, fitting the sea surface wind speed extraction model; applying the nonlinear quadratic function to fit the data f _f and s _f to obtain the sea surface wind speed prediction model:

Through data experimental verification, the quadratic function coefficients are finally obtained

β=325.9, δ=637.6.

Step 3.3, test the sea surface wind speed model; select the variance function SSE as the error test index for the model application test data, where SSE is the error square sum of the sea surface wind speed and the test wind speed obtained by inputting the mean value of the test radar echo intensity to the model, and the calculation formula is as follows:

in,

is the weighting coefficient, m is the number of data, s _i is the measured sea surface wind speed,

The sea surface wind speed extracted for the model. The closer the SSE is to 0, the more accurate the model and the higher the accuracy of the sea surface wind speed inversion.

Compared with the traditional method of curve fitting to extract the sea surface wind speed, the advantages of the present invention are:

1. A K-means clustering algorithm is designed to classify the data of the influencing factors of sea surface wind speed, and obtain heterogeneous data, and remove the influence of interference data on the sea surface model;

2. Apply a nonlinear quadratic function to remove heterogeneous data to determine the extraction model of sea surface wind speed to improve the extraction accuracy of sea surface wind speed;

3. The designed K-means clustering algorithm uses Euclidean distance to determine the distance of data points, and uses the distance mean as the determination condition for updating the centroid position, which has the advantages of fast convergence speed and good clustering effect;

4. The fitted model selects the variance function SSE as the error inspection index, which improves the accuracy of the wind speed extraction model and the inversion accuracy of the algorithm in engineering.

5. The model is obtained by training the measured X-band marine radar echo image, sea surface wind speed and sea surface wind speed, and has strong engineering applicability.

Description of drawings:

Fig. 1 is the specific embodiment flow chart of the present invention;

Figure 2 equipment acquisition and image sequence diagram;

Figure 3a is the radar map before filtering;

Figure 3b is the radar chart after filtering;

Figure 4 shows the relationship between sea surface wind speed and echo intensity;

Figure 5 is the distribution curve of the number of centroids and the sum of squares of errors;

Figure 6 is the K-means clustering distribution result of the influencing factors of sea surface wind speed;

Fig. 7 is the K-means clustering algorithm wind speed model fitting curve;

Fig. 8 is the fitting curve of the exponential function model wind speed model;

Figure 9 is a comparison result of the inversion results of the two groups of algorithms and the measured sea surface wind speed;

Figure 10 is a comparison chart of the error between the inversion results of the two groups of algorithms and the measured sea surface wind speed;

Figure 11 is a graph of the error statistics between the inversion results of the two groups of algorithms and the measured sea surface wind speed;

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. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The flowchart of the specific embodiment of the present invention is shown in Figure 1, which is divided into four major blocks: marine radar image preprocessing, radar data classification based on K-means clustering algorithm, sea surface wind speed extraction model determination, and sea surface wind speed information extraction. The specific implementation steps are divided into sixteen steps. The first to third steps are data preprocessing; the fourth to eleventh steps are radar data classification based on K-means clustering algorithm; the twelfth to tenth steps The fifth step is to determine the sea surface wind speed extraction model; the sixteenth step is to extract and analyze the sea surface wind speed information. Specific steps are as follows:

The first step is to collect 1722 sets of marine radar images from October 22-29, November 13-21, December 14-26, January 1-10, 2010 by self-made wave monitoring equipment Sequence, each group of image sequences contains 32 marine radar images, and the time for each image is 2.5s. The wind meters at the same position synchronously collect the corresponding sea surface wind direction information θ _w and wind speed information U _w . The equipment collection and image sequence are as follows: shown in Figure 2.

In the second step, for any pixel point of the radar image, the strength of the echo signal directly received by the radar is a voltage value of 0 to 2.5 volts. Therefore, the present invention returns the echo intensity value whose voltage value of the pixel point is less than 0.3V (according to the storage protocol, the echo intensity value after linear normalization of 0.3V voltage is 983) to zero intensity, and calculates the zero intensity pixel point. proportion. The formula for calculating the zero intensity percentage is as follows:

Among them, n is the total number of pixel points in the radar image, and n ₀ is the number of pixel points in the radar image whose echo intensity values return to zero.

Statistical analysis of the zero-intensity percentages of 332 groups in no rain, light rain, and heavy rain shows that under rainfall conditions, rain scattering has a greater impact on the radar, resulting in fewer invalid signals and a smaller zero-intensity percentage. In the end, the zero-intensity percentage is 0.542 when it is not raining, and the average zero-intensity percentage is 0.207 when it is raining. Therefore, the present invention determines radar images with zero intensity percentage lower than 0.207 as images with severe rainfall interference, and directly rejects them, and other images are images without rainfall interference, which are used for sea surface wind speed information extraction. Finally, 100 sets of images were removed by this method, and 1622 sets of images were reserved for subsequent research on sea surface wind speed information extraction technology.

The third step is to perform median filtering on the marine radar image g(x, y) after rainfall identification to suppress the influence of the same frequency signal on the extraction of sea surface wind speed. A 2D nonlinear smooth median filter of 3×3 template is applied to each radar image in the marine radar image sequence, and the filtered image gray value f'(x,y) is:

f'(x,y)＝median{g(x-i,y-j),(i,j)∈W} (2)

In formula (1), f(x, y) is the echo intensity value of the radar image; f'(x, y) is the gray value after filtering, and (i, j) is the 8 pixel points adjacent to the center of the template W . W is the template window, as follows:

11	11	11
11	11	11
11	11	11

The center of the median filter W is coincident with the center of the image, and the echo intensity value of the surrounding 8 adjacent pixel points is compared, and the middle value of the echo intensity is selected to update the echo intensity value of the image. The template is in step unit 1. Traverse the polar coordinate marine radar image, and finally obtain the marine radar image after median filtering. Figure 3a is the radar image before filtering, and Figure 3b is the radar image after filtering.

The fourth step is normalized data processing of the average value of radar echo intensity;

Select an appropriate part of the radar image for the preprocessed marine radar image, and perform normalized mapping along the x and y axes to obtain the average value of the radar image f' _i :

The fifth step is to perform normalized data processing on the image signal-to-noise ratio information;

Obtain the signal-to-noise ratio r _{t of the} selected radar image, normalize it with the radar image time series, and obtain the normalized value R _{i of the} sea state information:

in

is the noise spectrum other than the wave signal in the radar image.

The sixth step is to perform normalized data processing on the sea surface wind direction and wind speed information;

The sea surface wind direction information d _i and the sea surface wind speed information s _i collected by the anemometer are normalized according to the radar image sequence, and the normalized values D _i and S _{i of the} sea surface wind direction and wind speed information are obtained:

A seventh step, a third and fourth data set 1622 obtained in five steps S _{_{_i,}} R _i, D _i and F _i is divided into two portions, a data set for 1081 K-means clustering algorithm is trained surface wind speed experience Model, the remaining 541 sets of data are used for the test of the sea surface wind speed extraction model.

An eighth step, the statistical relationship between the wind speed data group 1081 and the echo intensity S _i F _i, as shown, can be seen in FIG. Surface wind speed and the echo intensity is proportional to 4, it can be seen with the radar image SSW The echo signals are closely correlated, and the sea surface wind field information can be retrieved using the marine radar image.

The relationship between the wind speed and the effective wave height of fully grown wind waves in deep water proposed by Wilson is as follows:

Among them, g is the acceleration of gravity, H _sw is the effective wave height of fully grown wind waves, and U is the wind speed. The above formula shows that the sea surface wind field and sea state information are closely related, and the signal-to-noise ratio of the radar image is proportional to the sea wave. Therefore, the present invention uses the echo intensity data F _i , the signal-to-noise ratio information R _i , and the sea surface wind speed information D _{i to} form a data set, as a set of factors affecting the sea surface wind speed, as follows:

Ω _i ={F _i ,R _i ,D _i } (8)

The ninth step is to divide the data points according to the initialized centroid;

① Initialize the cluster centroids, and randomly select _{K data points in the Ω i} region as the initialized centroids.

(2) After determining the centroids of the K sea surface wind speed influencing factors, find the data points closest to the centroids _{in the dataset Ω i to form clusters.} Here, the Euclidean distance is used for measurement, and _{the data points X i} (x ₁ ,x ₂ ,x ₃ ) of the characteristics of all the factors affecting the sea surface wind speed in _{Ω i} are calculated and the selected K centroids C _k (c ₁ ,c ₂ ,c ₃ ), the formula is as follows:

The tenth step, update the cluster centroids; for each Τ _k

According to the updated centroid, the Euclidean distance between the data point and the centroid is recalculated according to formula (9), and a new cluster is formed at the same time.

④The judgment basis for the centroid to stop updating;

||C _k -C _j ||＜＜γ (11)

Among them, γ=0.1, when the above conditions are met, it means that the centroids tend to converge, and the classification algorithm is terminated; if the above conditions are not met, repeat steps 8 and 9 until formula (11) is satisfied, and the cluster centroids C _f ( f = 1,2, ... f), and each cluster centroid corresponding data set Τ _f. The relationship between the number of centroids and the sum of squares of errors is obtained through experiments. As shown in Figure 5, when the number of cluster points is 5, it is the turning point of the sum of squares of errors, from which it decreases slowly, and the number of clusters is 5.

The eleventh step is to remove the radar data of heterogeneous data; the K-means clustering distribution results of the factors affecting the sea surface wind speed according to the number of clusters are shown in Figure 6, and the centroid with the farthest centroid position relative to other centroids is determined as heterogeneous The centroid C _d , all the data points in the region where it is located are also determined as heterogeneous data sets Τ _d , such as the cluster set corresponding to the green data in Fig. 6 . Removing the heterogeneous data in a data set Ω _i, S _i the simultaneous removal of heterogeneous data corresponding to a position of surface wind speed S _d, to give the final removal of heterogeneous radar data _{_{_{F f, R f, D f}}} , S f:

Ω _f ={F _f ,R _f ,D _f }={Ω _i -Τ _d } S _f ={S _i -S _d } (12)

The twelfth step, step 3.1, the sea surface wind speed extraction model is determined; first, the _{original characteristics of F f} _{, S f} data must be released to obtain the corresponding radar image echo intensity average f _f , and the training sea surface wind speed information s _f :

_{_{f f = F f * max (}} f i '), s f = S f * max (s i) (13)

The thirteenth step, fitting the sea surface wind speed extraction model; applying the nonlinear quadratic function to fit the data f _f and s _f to obtain the sea surface wind speed prediction model:

Among them, the quadratic function coefficient

is -9, β is 325.9, δ is -637.6, and the fitting curve is shown in Figure 7.

The fourteenth step, test the sea surface wind speed model; select the variance function SSE as the error test index for the model application test data, where SSE is the sum of squares of errors between the sea surface wind speed and the test wind speed obtained by inputting the mean value of the test radar echo intensity to the model, the calculation formula as follows:

in,

The sea surface wind speed extracted for the model. After experimental calculation, it is found that the SSE of the training data is 0.44, which is close to 0, indicating that the sea surface wind speed extraction model is accurate and can be used for engineering applications.

The fifteenth step, applying the exponential function relationship between the radar echo intensity and the sea surface wind speed proposed by Dankert, establishes the sea surface wind speed model as:

Among them, F _i is the radar echo intensity, S _i is the sea surface wind speed information, a, b, and c are the function coefficients, which are -0.7, -0.5, and 1.7, respectively. The fitting curve is shown in Figure 8. After experimental calculation, it is obtained that the SSE of the training data is 2.765, which is greater than the error function index of the algorithm of the present invention.

In the sixteenth step, the K-means clustering algorithm sea surface wind speed model and the exponential function sea surface wind speed model designed by the present invention are respectively applied to 541 groups of data, and the comparison result between the two groups of results and the measured sea surface wind speed is shown in Figure 9. It can be directly seen from Figure 9 that the sea surface wind speed model obtained by the K-means clustering algorithm is more consistent with the measured wind speed information, especially when the sea surface wind and rain are 15m/s, most of the wind speed information extracted by the exponential function sea surface wind speed model There is a problem that the wind speed is less than the measured wind speed.

Through the calculation of the experimental results, the inversion wind speed results of the two models and the measured wind speed statistical results are shown in Table 1. The correlation coefficient between the wind speed inversion results of the present invention and the measured wind direction reaches 0.99, the standard deviation is 0.38m/s, and the deviation is -0.04. Engineering requirements, and the results are completely superior to the exponential function inversion results, the inversion accuracy is improved by 77%.

Table 1 Error statistics of sea surface wind speed

The error comparison between the inversion results of the K-means clustering algorithm model and the inversion results of the exponential function model and the real value is shown in Figure 10. It can be seen from Figure 10 that the error range of the inversion results of the K-means clustering algorithm model is generally between -1 and +1m/s, while the error range of the exponential function model inversion results is between -4 and +6m/s. time, indicating that the K-means clustering algorithm model inversion results are more accurate. The error statistics of the results of the two algorithms are shown in Figure 11. It can be seen that the error range of the K-means clustering algorithm model inversion results is smaller, and about 50% of the data errors are concentrated in -0.1 ~ 0.1; while the exponential function The error range of the model inversion results is more scattered, and 62% of the data errors are concentrated in -1～1m/s. It can be concluded that the K-means clustering algorithm model inversion results are more accurate and stable than the exponential function model inversion results.

The above descriptions are only embodiments of the present invention, and common knowledge such as well-known specific structures and characteristics in the solution are not described too much here. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, but that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments are to be regarded in all respects as illustrative and not restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, which are therefore intended to fall within the scope of the appended claims. All changes within the meaning and range of the equivalents of , are included in the present invention. Any reference signs in the claims shall not be construed as limiting the involved claim.

Claims

A sea surface wind speed method based on K-means clustering algorithm, characterized in that the method is based on K-means clustering algorithm, and its implementation includes radar image data preprocessing, radar data classification based on K-means clustering algorithm , the determination of the sea surface wind speed extraction model and the extraction of sea surface wind speed information. The specific inversion steps are as follows:

Step 1, radar image data preprocessing: use the marine radar monitoring system to collect the sea surface radar image sequence data, synchronously use the wind meter to collect the synchronous sea surface wind direction and wind speed information, and apply the zero intensity percentage (ZPP) to the radar image sequence. Image data is identified and eliminated; for images with less interference from rain and snow, image median filtering is applied to suppress the interference of noise and co-frequency signals on the extraction of sea surface wind direction;

Step 2, data classification based on K-means clustering algorithm: First, perform data normalization on the radar image echo intensity, sea surface wind direction information, sea surface wind speed information and the calculated image signal-to-noise ratio, so that the data are in the same coordinate. Second, the K-means clustering algorithm is used to classify the data according to the Euclidean distance to the radar image echo intensity, sea surface wind direction information and image signal-to-noise ratio data, and the centroid distance error is used as the judgment basis to obtain heterogeneous data; finally , remove the information data corresponding to heterogeneous data from both the radar data and the sea surface wind field information data, and obtain the cluster data of the radar data and the sea surface wind field information;

Step 3, determining the sea surface wind speed extraction model: using the clustered radar data and the sea surface wind speed data to perform nonlinear quadratic fitting on the sea surface wind speed to obtain the sea surface wind speed extraction model, and applying SSE to verify the accuracy of the model;

Step 4, extraction of sea surface wind speed information: select some images of the test marine radar image, perform normalized mapping on them, and input them into the sea surface wind speed extraction model to obtain sea surface wind speed information.
A kind of sea surface wind speed method based on K-means clustering algorithm according to claim 1, is characterized in that:

The step 2 of the sea surface wind speed inversion includes the following steps:

Step 2.1, normalized data processing of the average value of radar echo intensity, sea surface wind direction, wind speed information and image signal-to-noise ratio information;

①Select an appropriate part of the radar image for the preprocessed marine radar image, and perform normalized mapping along the x and y axes to obtain the average value of the radar image f' i :

Wherein, f (x, y) is the radar image intensity values selected, N x, N y is the selected image along the x, y number of pixels, i is the corresponding radar image data; of f 'i is normalized in radar Image normalization value F i :

② Obtain the signal-to-noise ratio r t of the selected radar image, normalize it with the radar image time series, and obtain the normalized value R i of the sea state information:

in
is the corrected ocean wave spectrum of the two-dimensional wavenumber spectrum,
is the noise spectrum other than the radar image wave signal;

③ Normalize the sea surface wind direction information d i and the sea surface wind speed information s i collected by the anemometer according to the radar image sequence, and obtain the normalized values D i and S i of the sea surface wind direction and wind speed information:

Step 2.2, radar data classification based on K-means clustering algorithm;

① Initialize K initial cluster centroids;

Obtained in step 2.1 F i, R i, D i and S i of all the data divided into two parts; the data model for determination of F i, R i, D i consisting of a set of data as a set of surface wind speed factors ;

Ω i ={F i ,R i ,D i } (5)

Initialize the cluster centroids, and randomly select K data points in the Ω i region as the initialized centroids;

② Divide the data points according to the initialized centroid;

In determining the K surface wind speed affected by the quality of the heart, to find the distance from the centroid of the nearest data point in the data set Ω i, thereby forming clusters; Factor for all data points affect the calculation of surface wind speed Ω i of X i (x 1 ,x 2 ,x 3 ) and the distance between the selected K centroids C k (c 1 ,c 2 ,c 3 ), the formula is as follows:

After each point finds the closest centroid, it belongs to the cluster, and the data set Ω i is divided into K sub-region spaces Τ k ;

③ Update the cluster centroids;

for each Τ k
The mean value is used as the centroid of the next update, and the calculation formula is as follows:

According to the updated centroid, the Euclidean distance between the data point and the centroid is recalculated according to formula (6), and a new cluster is formed at the same time.

④The judgment basis for the centroid to stop updating;

According to the distance between the original centroid C k and the updated centroid C j , it is determined whether the centroid needs to be further updated. The judgment conditions are as follows:

||C k -C j ||＜＜γ (8)

When the above conditions are met, it means that the centroids tend to converge, and the classification algorithm is terminated; if the above conditions are not met, repeat steps 2.3 to 2.5 until formula (8) is met, and the cluster centroids C f (f=1,2, ... f), and each cluster centroid corresponding data set Τ f;

Step 2.3 remove radar data of heterogeneous data;

According to the obtained cluster distribution of the influencing factors of sea surface wind speed, the centroid whose centroid position is farthest from other centroids is determined as the heterogeneous centroid C d , and all the data points in its area are also determined as the heterogeneous data set Τ d , and the data is removed. heterogeneous data sets in Ω i, S i the simultaneous removal of heterogeneous data corresponding to a position of surface wind speed S d, to give the final removal of heterogeneous radar data F f, R f, D f , S f:

Ω f = {F f , R f , D f } = {Ω i -Τ d } S f ={S i -S d } (9)

Step 3.1, the sea surface wind speed extraction model is determined; first, the original characteristics of F f , S f data must be released to obtain the corresponding radar image echo intensity average value f f , and the training sea surface wind speed information s f :

f f = F f * max ( f i '), s f = S f * max (s i) (10)

Step 3.2, fitting the sea surface wind speed extraction model; applying the nonlinear quadratic function to fit the data f f and s f to obtain the sea surface wind speed prediction model:

Step 3.3, test the sea surface wind speed model; select the variance function SSE as the error test index for the model application test data, and the calculation formula is as follows:

Among them, ω i is the weighting coefficient, s i is the measured sea surface wind speed,
The sea surface wind speed extracted for the model; the closer the SSE is to 0, the more accurate the model and the higher the sea surface wind speed inversion accuracy.
A method for sea surface wind speed based on K-means clustering algorithm according to claim 2, characterized in that, in the step 2.2, initializing K initial cluster centroids, F i , R i , D i and S i The two parts are: one part is used for the determination of the sea surface wind speed model based on the K-means clustering algorithm, and the other part is used for the data testing of the model.
A kind of sea surface wind speed method based on K-means clustering algorithm according to claim 2, it is characterized in that, in described step 2.2, according to the initialization centroid division data point, all data points X i (x The distance between 1 , x 2 , x 3 ) and the selected K centroids C k (c 1 , c 2 , c 3 ) is the Euclidean distance.
The sea surface wind speed method based on the K-means clustering algorithm according to claim 2, wherein the centroid distance error judgment limit condition γ=0.1, and the number of clusters obtained is 5.
A sea surface wind speed method based on K-means clustering algorithm according to claim 2, characterized in that, in the step 3.2, the nonlinear quadratic function fits the data f f and s f , and the two Secondary function coefficients
is -9, beta is 325.9, and delta is -637.6.
A sea surface wind speed method based on K-means clustering algorithm according to claim 2, wherein, in the step 3.3, in the calculation of the test error test index of the sea surface wind speed model, the SSE is to input the test radar feedback to the model. The sum of squares of errors between the sea surface wind speed obtained from the mean wave intensity and the test wind speed.
A sea surface wind speed method based on K-means clustering algorithm according to claim 2, wherein, in the step 3.3, in the calculation of the test error test index of the sea surface wind speed model, the coefficient of the variance function SSE calculation formula
where m is the number of data.