CN114299377A - A vortex identification method and device based on width learning - Google Patents

Abstract

The invention discloses a vortex identification method and device based on width learning. The method includes: acquiring sea surface flow field data of a target area, and determining the sea-air interface of the target area according to the sea surface flow field data. flow field characteristics; judging the vortex distribution in the target area to determine the vortex distribution result of the target area; after fitting the vortex distribution result through the quadratic surface equation, determine the three-dimensional structure type of the vortex ; After learning the characteristics of the identified vortices by using the width learning method, predict the offshore sub-mesoscale vortices according to the flow field characteristics in the target area, and obtain the vortex identification result. The invention improves the accuracy of prediction and can be widely used in the technical field of data processing.

Description

A vortex identification method and device based on width learning

technical field

The invention relates to the technical field of data processing, in particular to a vortex identification method and device based on width learning.

Background technique

Eddy plays a non-negligible role in the transportation and distribution of global ocean matter and energy. Therefore, the identification and prediction of eddies is the frontier hotspot of current marine scientific research. However, due to the uncertainty in the location and time of sub-mesoscale eddies, their relatively small scales and the cost of large-scale on-site observation of the ocean Due to factors such as high cost, sub-mesoscale eddies observed in situ are very rare. Therefore, the exploration and innovation of identification and prediction methods of sub-mesoscale eddies are of great significance to the development of marine scientific research.

In the existing vortex identification and prediction methods, they can be divided into Euler method and Lagrangian method according to different categories of data, and the corresponding data types are Euler data and Lagrangian data respectively. Euler data refers to snapshot data or space field data at a moment, and Lagrangian data refers to the trajectory data of water masses or matter particles. Included in the Euler method are: OW parameter method, WA algorithm, VG algorithm, detection method based on sea surface temperature data, in addition to SAR image detection method for 3D eddy and aqua satellite detection method; Among the Longi methods, there are methods based on sea surface drifting buoy detection.

Although a variety of methods have been used to detect ocean eddies, none of them is suitable for eddies of various scales, and can accurately determine the physical quantities such as the position and structural scale of each vortex. These methods Each has its own advantages and disadvantages.

Although the OW algorithm is widely used, it still has three defects. The first is that it is difficult to determine the optimal threshold of the physical parameter W; the second is that the derivation process of the physical parameters will bring some noise terms, which will increase the false detection rate of the vortex; the third is that the physical standard will lead to the failure of the vortex detection. Or underestimate the size of the vortex size.

The VG algorithm gives four constraints, and at the same time, sensitivity experiments are needed to determine the parameters a and b. When the search area is small, the number of velocity minima points will increase, so the number of points used for the fourth constraint detection will increase the probability of misidentifying the vortex center. At the same time, if the vortex scale is small and close to the island or land, the velocity minimum point will be very close to the land, making it difficult to distinguish, so small eddies close to the coastline or between islands are easily missed. Also, a sleeve that is elongated or about to fall off completely can be misdetected as a vortex.

When using sea surface temperature data to detect vortices, only the thermal wind velocity vector is used instead of the velocity vector, and the detection method is still to screen the vortex center with four constraints, so the advantages and disadvantages of this detection method are similar to the VG algorithm.

When eddy detection is performed based on the trajectory of the sea-mark drifting buoy, the trajectory of the buoy in the actual ocean may be very complicated, and it is not an easy task to unambiguously identify all the eddies through the buoy. At the same time, if the background flow rate is greater than the tangential velocity of the vortex, the buoy will not form a loop after encountering the vortex, but a curve. At this time, it is necessary to remove the background flow field and reconstruct the Lagrangian trajectory, which will undoubtedly increase the Probing workload. Furthermore, the buoy only forms a loop inside the vortex, away from the edge of the vortex, so the size of the vortex estimated by the buoy data is small.

When using SAR images to detect sea surface vortices, only when the sea surface wind is too strong and the wind speed is too high, the sea surface wave action is too strong, so the sea surface is too rough and tends to be uniform, and the eddies cannot be clearly displayed on the SAR image. And the detection process requires certain conditions to obtain a clear SAR image, such as the convergence direction of the vortex, the direction of the wind-generated surface wave, and the radar viewing direction will all affect the clarity of the image.

If an aqua satellite is used to detect eddies, since the optical sensor will be affected by the cloud layer, effective ocean water color data cannot be obtained in the cloudy area. At this time, the data of multiple sensors needs to be fused, which will increase the detection cost. and increase in measurement error. In addition, there are few studies on the identification of eddies through water color remote sensing, and more research is on the influence of eddies on the distribution of chlorophyll on the sea surface. Therefore, this method of detecting eddies needs further exploration and research.

SUMMARY OF THE INVENTION

In view of this, embodiments of the present invention provide a method and device for vortex identification based on width learning with high accuracy.

One aspect of the present invention provides a width learning-based vortex identification method, comprising:

Acquire the sea surface flow field data of the target area, and determine the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data;

Perform vortex distribution judgment on the target area, and determine the vortex distribution result of the target area;

After fitting the vortex distribution result by the quadratic surface equation, the three-dimensional structure type of the vortex is determined;

After the features of the identified vortices are learned by using the width learning method, the offshore sub-mesoscale vortices are predicted according to the flow field features in the target area, and the vortex identification results are obtained.

Optionally, obtaining the sea surface flow field data of the target area, and determining the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data, including:

Collect the sea surface flow field data of the target area through high-frequency ground wave radar;

Data analysis is performed on the sea surface flow field data to obtain the flow field characteristics of the target area.

Optionally, performing vortex distribution judgment on the target area and determining a vortex distribution result of the target area includes:

Determine the air pressure feature and wind speed feature of the upper interface of the ocean vortex according to the prior knowledge, and perform a first analysis on the air pressure feature and the wind speed feature according to the satellite cloud image and the wind speed field image;

Converting the data collected by the high-frequency ground wave radar into the velocity vector field of the sea surface flow field, and performing a second analysis on the velocity vector field of the sea surface flow field through the VG algorithm;

acquiring the synchronous abnormal data of the sea surface temperature image in the infrared remote sensing image, and performing a third analysis on the sea surface temperature image;

According to the result of the first analysis, the result of the second analysis and the result of the third analysis, a vortex distribution result of the target area is determined.

Optionally, after the vortex distribution result is fitted by the quadratic surface equation, the three-dimensional structure type of the vortex is determined, including:

Convert the sea surface flow field data obtained by the high-frequency ground wave radar into a regional field velocity vector, and perform vortex initial identification on the regional field velocity vector;

Find an area with abnormal temperature in the remote sensing image, identify the position of the vortex based on the result of the initial identification of the vortex and the VG algorithm, and judge whether there is a vortex in the area;

After judging the existence of a vortex, take the vortex surface layer as the starting point, and layer down according to the preset interval distance to determine the flow velocity data of each vertical layer;

According to the direction of the velocity vector of each layer, find out whether there is a vortex with the same polarity as the surface layer in each layer;

If a vortex with the same polarity as the surface layer is found, the boundary of the vortex, the flow velocity at the center of the vortex, and the radius of the vortex are detected according to the shape of the velocity vector;

Corresponding coordinate systems are established at each layer respectively, and the boundary curve equation is obtained according to the vortex radius and the boundary fitting of the vortex;

The boundary surface equation of the three-dimensional vortex is obtained by fitting the boundary curve equation of each layer. The boundary surface equation is used to characterize the shape of the three-dimensional vortex, and the three-dimensional vortex shape includes a hyperbolic vortex and a parabolic vortex. ;

The significance test was carried out on the boundary surface equation of the fitted three-dimensional vortex, and the type of quadratic surface with the best test result was taken as the three-dimensional structure type of the vortex.

Optionally, after the feature of the identified vortex is learned by using the width learning method, prediction of the offshore sub-mesoscale vortex is performed according to the flow field feature in the target area, and the vortex identification result is obtained, including: :

The vortex center velocity, vortex radius, vortex surface shape, and infrared remote sensing image extraction features are used as input data;

dividing the input data into a training set and a validation set;

performing feature mapping on the training set, generating feature nodes, and obtaining an intermediate layer training matrix according to the feature nodes;

Performing nonlinear transformation processing on the feature node, generating an enhanced node, and obtaining an intermediate layer verification matrix according to the enhanced node;

splicing the feature node and the enhancement node to obtain a hidden layer;

output a predicted value according to the hidden layer, the middle layer training matrix and the middle layer verification matrix;

The vortex identification result is determined according to the predicted value.

Another aspect of the embodiments of the present invention also provides a width learning-based vortex identification device, including:

The first module is used to obtain the sea surface flow field data of the target area, and determine the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data;

The second module is used for judging the vortex distribution of the target area, and determining the vortex distribution result of the target area;

The third module is used to determine the three-dimensional structure type of the vortex after fitting the vortex distribution result through the quadratic surface equation;

The fourth module is used to predict the offshore sub-mesoscale eddies according to the flow field characteristics in the target area after learning the features of the identified vortices by using the width learning method, and obtain the vortex identification result.

Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;

the memory is used to store programs;

The processor executes the program to implement the method as described above.

Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, where the storage medium stores a program,

The program is executed by the processor to implement the method as previously described.

Another aspect of the embodiments of the present invention also provides a computer program product, including a computer program, which implements the aforementioned method when the computer program is executed by a processor.

The embodiment of the present invention obtains the sea surface flow field data of the target area, and determines the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data; and judges the vortex distribution in the target area, Determine the vortex distribution result of the target area; after the vortex distribution result is fitted by the quadratic surface equation, determine the three-dimensional structure type of the vortex; use the width learning method to perform the characteristics of the identified vortex. After learning, predict the offshore sub-mesoscale vortex according to the flow field characteristics in the target area, and obtain the vortex identification result. The present invention improves the accuracy of prediction.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the drawings that are used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a flow chart of the overall steps provided by an embodiment of the present invention;

Fig. 2 is the modeling flow chart of the width learning provided by the embodiment of the present invention;

FIG. 3 is a flowchart of an application based on width learning provided by an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

In view of the problems existing in the prior art, one aspect of the present invention provides a vortex identification method based on width learning, including:

Acquire the sea surface flow field data of the target area, and determine the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data;

Perform vortex distribution judgment on the target area, and determine the vortex distribution result of the target area;

After fitting the vortex distribution result by the quadratic surface equation, the three-dimensional structure type of the vortex is determined;

After the features of the identified vortices are learned by using the width learning method, the offshore sub-mesoscale vortices are predicted according to the flow field features in the target area, and the vortex identification results are obtained.

Specifically, the basic idea of the present invention is summarized as follows: (1) obtain meteorological data, remote sensing observation data, fixed station data, etc. in the study area, and analyze the data; (2) carry out meteorological element feature identification and satellite infrared remote sensing according to the obtained data. Image recognition and ground wave radar observation of sea surface flow field characteristics analysis to obtain sub-mesoscale eddy flow field and temperature characteristics, and comprehensively identify the region and scope of the existence of vortices; (3) Based on quadratic surface equations, the identified Three-dimensional structure determination and classification of vortex; (4) vortex prediction based on width learning.

Among them, meteorological element feature identification refers to: analyzing the relationship between sub-mesoscale eddies and sea-air interaction. For most areas with active meso- and small-scale eddies (such as the eastern Pacific Ocean and the extension of the Kuroshio), there is a positive correlation between sea surface temperature and sea surface wind speed on time scales such as pentad, day, month, and season. At the same time, the corresponding precipitation and cloud cover will increase (decrease) over the ocean's warm (cold) vortex. For this specific sea-air interaction relationship, according to the abundant satellite cloud images, wind speed and precipitation data of weather stations, and observation data, it is determined whether there is a point where precipitation, cloud amount and wind speed are abnormal with the background field. If compared with the background field, the point has more rainfall and cloudiness, and at the same time the wind speed increases, it is preliminarily judged that there is a warm vortex at this point; There is a cold vortex at the point. Determine whether the three meteorological element characteristics have corresponding positive (negative) changes, and then determine whether there is a vortex in the place. At the same time, the results of the VG algorithm combined with the high-frequency ground wave radar data and the infrared remote sensing image temperature anomaly discrimination method are combined to make a comprehensive judgment.

Optionally, obtaining the sea surface flow field data of the target area, and determining the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data, including:

Collect the sea surface flow field data of the target area through high-frequency ground wave radar;

Data analysis is performed on the sea surface flow field data to obtain the flow field characteristics of the target area.

Optionally, performing vortex distribution judgment on the target area and determining a vortex distribution result of the target area includes:

Determine the air pressure feature and wind speed feature of the upper interface of the ocean vortex according to the prior knowledge, and perform a first analysis on the air pressure feature and the wind speed feature according to the satellite cloud image and the wind speed field image;

Converting the data collected by the high-frequency ground wave radar into the velocity vector field of the sea surface flow field, and performing a second analysis on the velocity vector field of the sea surface flow field through the VG algorithm;

acquiring the synchronous abnormal data of the sea surface temperature image in the infrared remote sensing image, and performing a third analysis on the sea surface temperature image;

According to the result of the first analysis, the result of the second analysis and the result of the third analysis, a vortex distribution result of the target area is determined.

It should be noted that, for the second analysis, the present invention uses the high-frequency ground wave radar to observe the characteristics of the sea surface flow field, and aims to use the new type of observation equipment shore-based high-frequency ground wave radar to conduct wind and wave fields in the study area. , Real-time monitoring of the flow field, the time resolution of the system can reach 1 hour, and the spatial resolution can reach <1 km, which can monitor the sub-mesoscale dynamic processes and phenomena. According to the velocity field data, the VG algorithm for the geometrical characteristics of the flow field is used to calculate the vortex characteristic quantity.

The VG algorithm gives four constraints, and at the same time, sensitivity experiments are needed to determine the parameters a and b. When the search area is small, the number of velocity minima points will increase, so the number of points used for the fourth constraint detection will increase the probability of misidentifying the vortex center. At the same time, if the vortex scale is small and close to the island or land, the velocity minimum point will be very close to the land, making it difficult to distinguish, so small eddies close to the coastline or between islands are easily missed. Also, a sleeve that is elongated or about to fall off completely can be misdetected as a vortex. Therefore, if the prior art only uses the VG algorithm for vortex prediction, it has the above-mentioned disadvantages. In the present invention, the first analysis, the second analysis and the third analysis are respectively performed to assist the prediction, which is more accurate.

For the third analysis, the aim is to use the infrared satellite remote sensing sea surface temperature data set (such as GHRSST, etc.) to use the feature extraction method to automatically detect and identify vortices. At the same time, according to the thermal wind algorithm (Sobel operator and SST 2-dimensional convolution of the matrix of the The parameters obtained from the ground-wave radar detection data and the satellite meteorological data set are verified by each other. In addition, the infrared remote sensing images of the vortices in the study area are retained to prepare for the input of width learning data later.

Optionally, after the vortex distribution result is fitted by the quadratic surface equation, the three-dimensional structure type of the vortex is determined, including:

Convert the sea surface flow field data obtained by the high-frequency ground wave radar into a regional field velocity vector, and perform vortex initial identification on the regional field velocity vector;

Find an area with abnormal temperature in the remote sensing image, identify the position of the vortex based on the result of the initial identification of the vortex and the VG algorithm, and judge whether there is a vortex in the area;

After judging the existence of a vortex, take the vortex surface layer as the starting point, and layer down according to the preset interval distance to determine the flow velocity data of each vertical layer;

According to the direction of the velocity vector of each layer, find out whether there is a vortex with the same polarity as the surface layer in each layer;

If a vortex with the same polarity as the surface layer is found, the boundary of the vortex, the flow velocity at the center of the vortex, and the radius of the vortex are detected according to the shape of the velocity vector;

Corresponding coordinate systems are established at each layer respectively, and the boundary curve equation is obtained according to the vortex radius and the boundary fitting of the vortex;

The boundary surface equation of the three-dimensional vortex is obtained by fitting the boundary curve equation of each layer. The boundary surface equation is used to characterize the shape of the three-dimensional vortex, and the three-dimensional vortex shape includes a hyperbolic vortex and a parabolic vortex. ;

The significance test was carried out on the boundary surface equation of the fitted three-dimensional vortex, and the type of quadratic surface with the best test result was taken as the three-dimensional structure type of the vortex.

It should be noted that the acquisition of the quadratic surface equation of the vortex according to the present invention specifically refers to the starting point of the surface layer of the vortex center determined by infrared remote sensing images, satellite meteorological data sets, and high-frequency ground wave radar data sets, The vortex boundary curve equation is fitted by equidistant layers downward and the inflection point is determined outward from the center, and then three-dimensional fitting is performed by the two-dimensional vortex boundary fitting equation of each layer to obtain its quadratic Surface equation. Surface equations include ellipsoid, paraboloid, hyperboloid, cone and other types, and their expressions are as follows:

(a,b,c为正数)Ellipsoid:

(a,b,c are positive numbers)

(p,q同号)Ellipse paraboloid:

(same sign as p,q)

(p,q同号)Hyperbolic paraboloid:

(same sign as p,q)

(a,b,c为正数)Single leaf hyperboloid:

(a,b,c are positive numbers)

(a,b,c为正数)Double leaf hyperboloid:

(a,b,c are positive numbers)

(a,b为正数)Ellipse cone:

(a,b are positive numbers)

In this embodiment, the above quadric surface expressions are fitted one by one, and then significance tests are performed respectively, and the quadric surface type with the best significance test result is used as the three-dimensional structure type of the vortex.

Among them, the three-dimensional structure determination refers to the quadratic surface equation obtained from the coordinate system established with the vortex center as the origin, and the vortex The three-dimensional structure of the spin is determined and classified.

Optionally, after using the width learning method to learn the characteristics of the identified vortices, predict the offshore sub-mesoscale vortices according to the flow field characteristics in the target area, and obtain a vortex identification result, including: :

The vortex center velocity, vortex radius, vortex surface shape, and infrared remote sensing image extraction features are used as input data;

dividing the input data into a training set and a validation set;

performing feature mapping on the training set, generating feature nodes, and obtaining an intermediate layer training matrix according to the feature nodes;

Performing nonlinear transformation processing on the feature node, generating an enhanced node, and obtaining an intermediate layer verification matrix according to the enhanced node;

splicing the feature node and the enhancement node to obtain a hidden layer;

output a predicted value according to the hidden layer, the middle layer training matrix and the middle layer verification matrix;

The vortex identification result is determined according to the predicted value.

It should be noted that the present invention adopts the width learning system to learn the identified vortices. For a large amount of data, the breadth learning system automatically learns the implicit structure and existing rules between the data, so as to make corresponding predictions for the newly input data. It takes advantage of two major advantages that are different from other machine learning systems, namely (1) based on the RVFL network and (2) simple structure. The invention takes the random vector function linking neural network as the carrier, and realizes the horizontal expansion of the design network by increasing the increment of neural nodes instead of the number of structural layers to achieve the purpose of prediction. This method takes the vortex center velocity, vortex radius, vortex surface shape, and infrared remote sensing image extraction features as input data, and the output data is the prediction result of vortex change radius, life span, moving path, etc. The model generates feature nodes and enhancement nodes through feature mapping and nonlinear transformation, which are jointly used as hidden layers to output results. At the same time, the input data is divided into training sets and validation sets to train and test the model.

Another aspect of the embodiments of the present invention also provides a width learning-based vortex identification device, including:

The first module is used to obtain the sea surface flow field data of the target area, and determine the flow field characteristics of the sea-air interface of the target area according to the sea surface flow field data;

The second module is used for judging the vortex distribution of the target area, and determining the vortex distribution result of the target area;

The third module is used to determine the three-dimensional structure type of the vortex after fitting the vortex distribution result through the quadratic surface equation;

The fourth module is used to predict the offshore sub-mesoscale eddies according to the flow field characteristics in the target area after learning the features of the identified vortices by using the width learning method, and obtain the vortex identification result.

Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;

the memory is used to store programs;

The processor executes the program to implement the method as described above.

Another aspect of the embodiments of the present invention further provides a computer-readable storage medium, where the storage medium stores a program,

The program is executed by the processor to implement the method as previously described.

Another aspect of the embodiments of the present invention also provides a computer program product, including a computer program, which implements the aforementioned method when the computer program is executed by a processor.

The specific implementation principle of the present invention will be described in detail below in conjunction with the accompanying drawings:

The embodiment of the present invention proposes a new type of vortex identification method, and the main innovative contents include the following parts: 1. Using the Broad Learning System to predict vortices, the relevant content can be understood first; 2. From the sea-air interface Start with features, consider the characteristics of air pressure and wind speed (the upper interface of the ocean vortex) to identify the vortex; 3. Combine the quadratic surface equation to make a priori determination of the vortex shape.

Specifically, the overall implementation steps of the present invention are shown in Figure 1:

Step 1: Obtain the high-frequency ground wave radar observation sea surface flow field data in the study area, and obtain the flow field characteristics of the sea-air interface in the study area by processing and analyzing the data;

As an emerging marine environment observation technology, high-frequency ground wave radar can detect the information of the sea surface flow field within the range of 200km from the coast by using the characteristics of small diffraction propagation attenuation on the conductive ocean surface compared with short waves (3-30MHZ). It has the advantages of over-the-horizon detection, large coverage, high detection accuracy, moderate cost, good real-time performance, not affected by bad weather and measured sea conditions, and can work around the clock. The basic principle is that a wave can be decomposed into a superposition of multiple simple sinusoidal wave train components with different amplitudes, periods, initial phases and propagation directions. All simple sine wave trains interact with high-frequency electromagnetic waves to produce scattering effects, but the contributions of different sine wave components are different. Only when the following two conditions are met, the radar will receive a strong echo: ① The wavelength is equal to half of the wavelength of the radio wave; ② only when the propagation direction is close to the radar or far away from the radar will Bragg resonance occur with the radio wave emitted by the radar. Therefore, this scheme uses high-frequency ground wave radar to collect regional ocean current flow field data information, which is beneficial to improve the accuracy of the data.

The second step: use the meteorological (wind speed field, satellite cloud image) element data of satellite remote sensing to indirectly identify the vortex, and then combine the VG algorithm to process the high-frequency ground wave radar data, and the automatic interpretation algorithm to analyze the infrared remote sensing image. Does the regional vortex exist?

1) The air pressure and wind speed characteristics of the upper interface of the ocean vortex are summarized in the existing research data, and comprehensive analysis is carried out according to the existing abundant satellite cloud images and wind speed field images. Compared with the large-scale (such as the North Pacific Basin) performance is the influence of the atmosphere on the ocean, that is, the negative correlation between ocean surface temperature and wind speed. The local area where the vortex exists mainly reflects the influence of the ocean on the atmosphere, that is, the ocean surface temperature has a significant positive correlation with the wind speed and cloud amount above the vortex. relationship, while water vapor content and precipitation also respond (although the correlation is smaller than for wind speed and cloud cover).

Changes in the stability of the atmospheric boundary layer, as well as convection (enhancement/suppression) and changes in water vapor supply are possible causes, according to available data. According to the vertical mixing mechanism of surface momentum, high SST (warm vortex) will make the atmosphere in the boundary layer unstable, and vertical mixing will be enhanced, causing a large momentum to be transmitted down in the middle and upper layers of the boundary layer, which will increase the sea surface wind speed. On the contrary, cold SST (cold vortex) It will increase the stability of the atmosphere and inhibit turbulent mixing.

2) Convert the high-frequency ground wave radar data into the velocity vector field of the sea surface flow field, and perform vortex identification and analysis according to the VG algorithm.

The VG algorithm defines a vortex based on the geometrical characteristics of the flow field: a vortex can be intuitively defined as a region where the velocity vector rotates clockwise or counterclockwise around a central point. Several current studies have pointed out some typical characteristics describing the vortex velocity field: the velocity near the vortex center is the smallest; the magnitude of the tangential velocity increases nearly linearly with the distance from the center point, and decays after reaching a maximum somewhere. The VG algorithm proposes four constraints corresponding to the definition of the vortex velocity field and the above characteristics, and the point satisfying all the constraints is defined as the vortex center.

3) Based on the SST anomaly in infrared remote sensing images, the feature extraction method in the automatic interpretation algorithm of vortex infrared remote sensing is used to extract image features and identify vortices.

Since the warm vortex and cold vortex have positive and negative seawater temperature anomalies centered on the vortex center, these anomalies can reach the ocean surface under ideal conditions, forming a synchronous anomaly of sea surface temperature SST (Sea Surface Temperature), so that they are carried by satellites. detected by infrared sensors. Therefore, vortices can be effectively identified based on infrared remote sensing images.

With the continuous increase of the amount of SST data, the inefficiency and disadvantage of visual interpretation methods have become increasingly prominent, and vortex infrared remote sensing has gradually moved from manual interpretation to automatic interpretation. Among them, the feature extraction method uses a specific algorithm to calculate the feature velocity from the continuous SST images, and locates and identifies the vortex in the velocity vector field, and uses as few new features as possible to maximize the inclusion of more effective vortex information. The feature extraction method can not only be used for eddy detection in satellite SST field sequence features, but also for advanced expression of large-scale geophysical data, high-precision SST data reconstruction, and mining and analysis of ocean big data.

Use the above three methods 1)-3) to identify vortices in the study area respectively, and comprehensively analyze the identification results to obtain the final vortex distribution results in the study area.

According to the existing data, the characteristics of ocean eddies are: the seawater flow inside the eddies can form a clockwise or counterclockwise rotating flow field, and cause the surface seawater to converge (upwelling) or diverge (downflowing), resulting in eddies The height of the sea surface at the center of the rotation decreases or increases. At the same time, for most of the sea areas with active eddies, there is an obvious positive correlation between the sea surface temperature and the wind speed and cloud amount above the eddies; the warm and cold vortices have positive and negative seawater temperature anomalies centered on the vortex center, respectively. , resulting in a decrease or increase in the air pressure on the surface of the vortex; if the system of the atmosphere above the sea surface temperature anomaly caused by the ocean vortex moves slowly, it is less likely to produce strong sea-air convection; The cyclonic atmospheric system moves faster, and the ocean vortex affects the atmosphere through a vertical mixing mechanism, which makes it easier to form a strong convergence and rise on the downstream side of the background wind of the warm vortex; at the same time, the warm vortex will increase the sea surface wind speed, and the cold vortex will The surface atmosphere becomes stable. According to the above characteristics of vortex surface temperature, air pressure and wind speed, it is considered to combine the velocity vector of the sea surface flow field obtained from the high-frequency ground wave radar data with the remote sensing image to identify the sub-mesoscale vortex.

Step 3: Fit the identified vortex shape with quadratic surface equation, and make a priori judgment on it;

1), convert the sea surface flow data obtained by the high-frequency ground wave radar into the regional field velocity vector, and use the sea surface velocity vector to carry out vortex identification;

2) Find an area with abnormal temperature in the remote sensing image, and make a comprehensive judgment on whether there is a vortex in the area in combination with the vortex position identified by the VG algorithm;

3), take the vortex surface layer as the starting point, layer down with a certain interval distance, and output the flow velocity data of each vertical layer from the result model;

4), according to the direction of the velocity vector of each layer, find out whether there is a vortex with the same polarity as the surface layer in each layer;

5) If the corresponding vortex can be found, the boundary of the layer of vortex, the flow velocity of the vortex center, and the vortex radius are detected according to the shape of the velocity vector. The vortex boundary is defined as the velocity field decreases from the center outwards. The connecting line to the inflection point that increases from the minimum value, the average distance between the inflection point and the center point of the vortex is defined as the radius of the vortex; if no corresponding vortex is found, it is considered that the maximum depth of the vortex is smaller than the layer depth;

6), establish a corresponding coordinate system in each layer, fit the boundary curve equation according to the measured vortex radius and boundary shape, and then fit the boundary surface equation of the three-dimensional vortex from the boundary curve equation of each layer, thus Determine the shape of the vortex (including hyperbolic vortex, parabolic vortex, etc.);

7), quadric surfaces can be divided into ellipsoid, paraboloid, hyperboloid, cone and other types, the respective expressions are as follows:

(a,b,c为正数)Ellipsoid:

(a,b,c are positive numbers)

(p,q同号)Ellipse paraboloid:

(same sign as p,q)

(p,q同号)Hyperbolic paraboloid:

(same sign as p,q)

(a,b,c为正数)Single leaf hyperboloid:

(a,b,c are positive numbers)

(a,b,c为正数)Double leaf hyperboloid:

(a,b,c are positive numbers)

(a,b为正数)Ellipse cone:

(a,b are positive numbers)

First fit the above quadric expressions one by one, and then perform significance tests respectively, and take the quadric type with the best significance test result as the three-dimensional structure type of the vortex.

Step 4: Use the width learning method to learn the characteristics of the identified eddies, and then predict the offshore sub-mesoscale eddies according to the flow field characteristics in the study area;

As a neural network structure that does not rely on deep structure, breadth learning has the advantages of fast operation speed and simple system structure compared with the currently widely used deep learning system. By using the width learning method, the invention can realize the training and learning of the known vortex data, and then realize the prediction of the data, such as predicting the diameter, lifespan, moving path and the like of the vortex.

As shown in FIG. 2 and FIG. 3 , the specific process of processing data in the width learning method according to the embodiment of the present invention is as follows:

1) The vortex center velocity, vortex radius, vortex surface shape, and infrared remote sensing image extraction features are used as input data. The hyperboloid and conical surface are set as 1, 2, 3, and 4 respectively, as the corresponding numerical input. The extraction features of the infrared remote sensing image are similar to the input method of the vortex surface shape.

2) Divide the input data into training sets Xtrain, Ytrain, validation sets Xtest, Ytest, and the width learning network output model can be expressed as Y=HW;

其中

为线性或非线性激活函数，W_ei和

分别为随机权重和偏置。将n组特征节点拼接为Zⁿ＝[Z₁，Z₂，…，Z_n]；3), the training set data Xtrain is generated through feature mapping to generate feature nodes, and the intermediate layer training matrix Htrain is obtained. Use Z _i to represent the i-th group of feature nodes containing q neurons, then we have

in

is a linear or nonlinear activation function, _Wei and

are random weights and biases, respectively. Splicing n groups of feature nodes into Z ⁿ =[Z ₁ , Z ₂ ,..., Z _n ];

其中ξ_j是非线性激活函数，W_hj和

分别是随机权重和偏置。将m组增强节点拼接为H^m＝[H₁，H₂，...，H_m]；4), the feature node is subjected to nonlinear transformation to generate an enhanced node, and an intermediate layer verification matrix H _test is obtained. Use H _j to represent the j group of enhanced nodes containing r neurons, then we have

where ξ _j is the nonlinear activation function, W _hj and

are random weights and biases, respectively. Concatenate m groups of enhancement nodes as H ^m =[H ₁ , H ₂ , . . . , H _m ];

5), splicing feature nodes and enhancement nodes together as a hidden layer;

6) The output of the hidden layer is obtained through the connection weight to obtain the final output, that is, the predicted value is obtained according to Y _test = H _test W, and the output results are the vortex change radius, life, moving path, etc.

To sum up, compared with the prior art, the embodiment of the present invention has the following advantages:

The vortex detection method not only identifies the sea surface vortex according to the vortex surface characteristics of the sea surface pressure map, but also obtains its three-dimensional structure and shape, and uses the numerical equation method to express its shape characteristics. The translation algorithm and the data results of the high-frequency ground wave radar are verified each other, and the vortex is comprehensively identified by the dynamic method, the velocity vector geometry method and the temperature anomaly feature based on the vortex. This method is not affected by the above-mentioned cloud cover, etc., but also avoids the misjudgment of vortex to a certain extent.

In addition, various existing vortex detection methods focus on the identification of vortices, and this method also uses the width learning method to learn the characteristics of the identified vortices, and then according to the characteristics of the flow field in the study area. The prediction of offshore sub-mesoscale eddies is of great significance to the study of eddies.

In some alternative implementations, the functions/operations noted in the block diagrams may occur out of the order noted in the operational diagrams. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may sometimes be executed in the reverse order, depending upon the functionality/operations involved. Furthermore, the embodiments presented and described in the flowcharts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and in which sub-operations described as part of larger operations are performed independently.

Furthermore, while the invention is described in the context of functional modules, it is to be understood that, unless stated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or or software modules, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be appreciated that a detailed discussion of the actual implementation of each module is not necessary to understand the present invention. Rather, given the attributes, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of such modules will be within the routine skill of the engineer. Accordingly, those skilled in the art, using ordinary skill, can implement the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are illustrative only and are not intended to limit the scope of the invention, which is to be determined by the appended claims along with their full scope of equivalents.

The functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

The logic and/or steps represented in flowcharts or otherwise described herein, for example, may be considered an ordered listing of executable instructions for implementing the logical functions, may be embodied in any computer-readable medium, For use with, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other system that can fetch instructions from and execute instructions from an instruction execution system, apparatus, or apparatus) or equipment. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport the program for use by or in connection with an instruction execution system, apparatus, or apparatus.

More specific examples (non-exhaustive list) of computer readable media include the following: electrical connections with one or more wiring (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), Read Only Memory (ROM), Erasable Editable Read Only Memory (EPROM or Flash Memory), Fiber Optic Devices, and Portable Compact Disc Read Only Memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program may be printed, as the paper or other medium may be optically scanned, for example, followed by editing, interpretation, or other suitable medium as necessary process to obtain the program electronically and then store it in computer memory.

It should be understood that various parts of the present invention may be implemented in hardware, software, firmware or a combination thereof. In the above-described embodiments, various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented by any one or a combination of the following techniques known in the art: Discrete logic circuits, application specific integrated circuits with suitable combinational logic gates, Programmable Gate Arrays (PGA), Field Programmable Gate Arrays (FPGA), etc.

In the description of this specification, description with reference to the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples", etc., mean specific features described in connection with the embodiment or example , structure, material or feature is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, The scope of the invention is defined by the claims and their equivalents.

The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments, and those skilled in the art can also make various equivalent deformations or replacements on the premise of not violating the spirit of the present invention, These equivalent modifications or substitutions are all included within the scope defined by the claims of the present application.

Claims (10)

1. A vortex identification method based on width learning is characterized by comprising the following steps:

acquiring the sea surface flow field data of a target area, and determining the flow field characteristics of a sea-air interface of the target area according to the sea surface flow field data;

carrying out vortex distribution judgment on the target area, and determining a vortex distribution result of the target area;

fitting the vortex distribution result through a quadric surface equation, and determining the three-dimensional structure type of the vortex;

and after learning the characteristics of the identified vortex by adopting a width learning method, predicting the offshore secondary mesoscale vortex according to the flow field characteristics in the target area to obtain a vortex identification result.

2. The method for vortex identification based on width learning according to claim 1, wherein the obtaining of the surface flow field data of the target area and the determining of the flow field characteristics of the sea-air interface of the target area according to the surface flow field data comprises:

collecting the sea surface flow field data of the target area through a high-frequency ground wave radar;

and performing data analysis on the sea surface flow field data to obtain the flow field characteristics of the target area.

3. The method for vortex identification based on width learning according to claim 1, wherein the determining vortex distribution of the target region and determining a vortex distribution result of the target region includes:

determining the air pressure characteristic and the wind speed characteristic of an interface on an ocean vortex according to prior knowledge, and performing first analysis on the air pressure characteristic and the wind speed characteristic according to a satellite cloud picture and a wind speed field image;

converting data acquired by a high-frequency ground wave radar into a sea surface flow field speed vector field, and performing second analysis on the sea surface flow field speed vector field through a VG algorithm;

acquiring synchronous abnormal data of a sea surface temperature image in an infrared remote sensing image, and performing third analysis on the sea surface temperature image;

and determining the vortex distribution result of the target area according to the result of the first analysis, the result of the second analysis and the result of the third analysis.

4. The method of claim 3, wherein the determining the three-dimensional structure type of the vortex after fitting the vortex distribution result by a quadric equation comprises:

converting sea surface flow field data acquired by a high-frequency ground wave radar into a regional field flow velocity vector, and performing vortex primary identification on the regional field flow velocity vector;

finding an area with abnormal temperature in the remote sensing image, identifying the vortex position by combining the initial vortex identification result and a VG algorithm, and judging whether the area has vortex;

after the vortex is judged to exist, taking the vortex surface layer as a starting point, layering downwards according to a preset spacing distance, and determining flow speed data of each layer in the vertical direction;

according to the direction of the flow velocity vector of each layer, whether vortex with the same polarity as that of the surface layer exists in each layer is searched;

if the vortex with the same polarity as the surface layer is found, detecting the boundary of the vortex, the central flow velocity of the vortex and the radius of the vortex according to the shape formed by the flow velocity vector;

establishing corresponding coordinate systems on each layer respectively, and fitting according to the vortex radius and the vortex boundary to obtain a boundary curve equation;

fitting according to a boundary curve equation of each layer to obtain a boundary surface equation of the three-dimensional vortex, wherein the boundary surface equation is used for representing the form of the three-dimensional vortex, and the form of the three-dimensional vortex comprises a hyperboloid vortex and a paraboloid vortex;

and (4) carrying out significance test on the boundary surface equation of the three-dimensional vortex obtained by fitting, and taking the quadric surface type with the best test result as the three-dimensional structure type of the vortex.

5. The method for vortex identification based on width learning according to claim 1, wherein after learning the features of the identified vortex by using the width learning method, predicting the offshore secondary mesoscale vortex according to the flow field features in the target region to obtain a vortex identification result, comprising:

extracting characteristics of vortex central flow velocity, vortex radius, vortex curved surface form and infrared remote sensing images to serve as input data;

dividing the input data into a training set and a validation set;

performing feature mapping on the training set to generate feature nodes, and obtaining an intermediate layer training matrix according to the feature nodes;

carrying out nonlinear transformation processing on the characteristic nodes to generate enhanced nodes, and obtaining an intermediate layer verification matrix according to the enhanced nodes;

splicing the characteristic nodes and the enhanced nodes to obtain a hidden layer;

outputting a predicted value according to the hidden layer, the middle layer training matrix and the middle layer verification matrix;

and determining the vortex identification result according to the predicted value.

6. The method of claim 5, wherein the vortex identification result includes, but is not limited to, a moving radius of a vortex, a life of the vortex, and a moving path of the vortex.

7. A vortex identification device based on width learning, comprising:

the device comprises a first module, a second module and a third module, wherein the first module is used for acquiring the sea surface flow field data of a target area and determining the flow field characteristics of a sea-air interface of the target area according to the sea surface flow field data;

the second module is used for judging vortex distribution of the target area and determining a vortex distribution result of the target area;

the third module is used for determining the three-dimensional structure type of the vortex after fitting the vortex distribution result through a quadric surface equation;

and the fourth module is used for predicting the offshore secondary mesoscale vortex according to the flow field characteristics in the target area after learning the characteristics of the identified vortex by adopting a width learning method, so as to obtain a vortex identification result.

8. An electronic device comprising a processor and a memory;

the memory is used for storing programs;

the processor executing the program realizes the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that the storage medium stores a program, which is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that the computer program realizes the method of any of claims 1 to 6 when executed by a processor.

Images