Vector diagram feature extraction method for slot ridge line identification

Abstract

The invention is applicable to the technical field of meteorological engineering, and provides a vector diagram feature extraction method for groove ridge line recognition, which comprises the following steps: firstly, acquiring contour line data and wind field data, then traversing all lines in the contour line data, when the number of points included in the current line is at least three, extracting geometric structure features of each point of the current line, and extracting the geometric structure features of each point of the current line; then extracting four wind field component fields in a preset range of each point in the current line, calculating wind direction angles and wind field types of the component fields, and finally constructing a feature matrix of the line according to geometric structure features, the wind direction angles and the wind field types of each point of the current line. Therefore, by utilizing the vectorization structure characteristics of the contour line and the wind field based on latitude and longitude coordinates, and the advantages that the vector diagram is small in size, accurate in coordinate and clear in boundary, the groove ridge line recognition efficiency and accuracy are jointly improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of meteorological engineering, and particularly relates to a vector graph feature extraction method for trough-ridge line identification. BACKGROUND

[0002] The trough line is a line connecting the maximum curvature points of the isohyet in the low-pressure trough, is the core area of cyclonic convergence (and upward motion), and is often accompanied by the development of a precipitation system in front of it; the ridge line is a line connecting the maximum anticyclonic curvature points of the isohyet in the high-pressure ridge, is the core area of anticyclonic divergence (and sinking motion), and is usually associated with clear and stable weather activities. As a core weather system in a weather map, it is drawn based on the isohyet and corresponding wind field on the 200-850 hPa (especially 500 hPa) isobaric surface, and the recognition accuracy directly affects the timeliness of the warning of disastrous weather.

[0003] The previous identification of trough-ridge lines mainly relies on the calculation of the curvature of the isohyet and the gradient calculation based on the height field grid data. However, in recent years, with the vigorous development of machine learning technology, CNN, LSTM, and other neural networks have mature technology implementation and various applications in the meteorological field. Therefore, related research has been carried out in the identification of trough-ridge lines in combination with machine learning technology, and certain achievements have been made. In the process of identifying trough-ridge lines using CNN convolutional neural networks, it is necessary to first locate the sub-image that may contain the trough-ridge line in the weather map, then establish a set of trough-ridge line classification (regression) model to identify the trough-ridge line of the sub-image, and further draw the trough-ridge line.

[0004] However, in the current technology, the trough-ridge line identification method based on pixel graphs has a series of problems such as blurred line boundaries, inaccurate feature positioning, strong dependence on resolution, and the like, resulting in low efficiency and accuracy of trough-ridge line identification. SUMMARY

[0005] The vector graph feature extraction method for trough-ridge line identification provided in the embodiments of the application can solve the above problems. The method takes advantage of the vectorization structure characteristics of the isohyet and wind field based on latitude and longitude coordinates, and the advantages of small volume, accurate line coordinates, and clear boundaries of vector graphs, thereby improving the recognition efficiency and accuracy of the machine learning model.

[0006] In a first aspect, the embodiments of the present application provide a vector map feature extraction method for ridge line identification, comprising the following steps: S1, obtaining contour line data and wind field data; S2, traversing all lines in the contour line data, and when the number of points contained in a current line is at least 3, extracting the geometric structure feature of each point of the current line; S3, extracting four wind field component fields in a preset range of each point in the current line, and calculating the wind direction angle and the wind field type of the component fields; and S4, constructing a feature matrix of the line according to the geometric structure feature, the wind direction angle and the wind field type of each point of the current line.

[0007] In a possible implementation form of the first aspect, the structure of the contour line data in S1 is contour={cloc,ctag}, cloc is a line list containing cn lines, and the line c∈{c1,c2,...c cn}, c is a longitude and latitude coordinate matrix with dimensions [cnc,2], where cnc represents the number of points constituting the cth line, and each point contains longitude and latitude data; ctag is a tag sequence containing cn potential height data; the structure of the wind field data in S1 is wind={wlon,wlat,u,v,w}, where wlon is an equidistant longitude coordinate sequence with a length of wnc; wlat is an equidistant latitude coordinate sequence with a length of wnr; u is a horizontal meridional component field, which is a matrix with dimensions [wnr,wnc]; v is a horizontal zonal component field, which is a matrix with dimensions [wnr,wnc]; and w is a vertical component field, which is a matrix with dimensions [wnr,wnc].

[0008] Optionally, in another possible implementation form of the first aspect, the current line includes a set of two-dimensional coordinate points c=[p1,p2,...,p cnc ], where p i =(x i ,y i ) represents a point on the current line, x i is the longitude, and y i is the latitude.

[0009] The first-order derivative and the second-order derivative of the point on the current line are calculated by the difference method.

[0010] The normalized curvature of each point is calculated based on the first-order derivative and the second-order derivative.

[0011] The concave-convex point detection based on local maximum curvature is performed according to the normalized curvature.

[0012] The tangent direction angle and the global direction consistency of the current line are calculated.

[0013] Optionally, in a further possible implementation manner of the first aspect, the points on the current line include an interior point and a boundary point, and the first derivative and the second derivative of the point on the current line are calculated by the difference method, specifically as follows:

[0014] The first derivative of the interior point is:

[0015]

[0016] The first derivative of the boundary point is:

[0017] The first point dx1=x2-x1, dy1=y2-y1

[0018] The last point dx cnc =x cnc -x cnc-1 , dy cnc =y cnc -y cnc-1

[0019] The second derivative of the interior point is:

[0020]

[0021] The second derivative of the boundary point is:

[0022] The first point ddx1=dx2-dx1, ddy1=dy2-dy1

[0023] The last point ddx cnc =dx cnc -dx cnc-1 , ddy cnc =dy cnc -dy cnc-1 .

[0024] Optionally, in a further possible implementation manner of the first aspect, the calculating the normalized curvature of each point based on the first derivative and the second derivative includes:

[0025] The curvature of each point is calculated as:

[0026]

[0027] Wherein ε=10 -6 is a parameter for preventing division by zero error;

[0028] The curvature of each point is normalized to obtain the normalized curvature of each point:

[0029]

[0030] Optionally, in a further possible implementation manner of the first aspect, the local maximum curvature based concave-convex point detection according to the standardized curvature comprises:

[0031] Performing convex point labeling:

[0032]

[0033] Performing concave point labeling:

[0034]

[0035] Performing boundary point processing:

[0036] cvex1 = ccave1 = cvex cnc = ccave cnc = 0.

[0037] Optionally, in a further possible implementation manner of the first aspect, the tangent direction angle and the global direction consistency are calculated as follows:

[0038] Tangent direction angle:

[0039] θ i = arctan2 (dy i , dx i ) i ∈ {1,... cnc}

[0040] Where -π < θ i ≤ π, arctan2 is a complete mapping function from coordinates to angles, which realizes the full quadrant coverage and boundary processing ability through two parameters;

[0041] Global direction consistency:

[0042]

[0043] Where cnc j represents the number of points of the jth line, dy i,j is the first order derivative of the latitude of the ith point of the jth line, dx i,j is the first order derivative of the longitude of the ith point of the jth line, θ global is the average of the tangent direction angles of all points in the line list cloc;

[0044] Δθ global,i = min (|θ i - θ global |, 2π - |θ i - θ global |) i ∈ {1,... cnc}

[0045] Where, Δθglobal,i the deviation between the tangent direction angle of the i-th point of the current line and θ global

[0046] Optionally, in a possible implementation of the first aspect, the four wind field component fields in the preset range of each point in the current line are extracted, and the wind direction angle and the wind field type of the component field are calculated, in the S3, including:

[0047] In the current line, the wind field component field of four grid points in the preset range around each point is extracted:

[0048] u i =[u i,1 ,u i,2 ,u i,3 ,u i,4 ],v i =[v i,1 ,v i,2 ,v i,3 ,v i,4 ],w i =[w i,1 ,w i,2 ,w i,3 ,w i,4 ]

[0049] The wind direction angle of each point is calculated:

[0050]

[0051] wherein

[0052] The wind field type of each point is calculated:

[0053]

[0054] Optionally, in a possible implementation of the first aspect, the feature matrix of the line is constructed according to the geometric structural features, the wind direction angle and the wind field type of each point in the current line, in the S4, including:

[0055] The feature vector of each point is constructed:

[0056]

[0057] The feature matrix of the line is constructed according to the feature vector of each point:

[0058]

[0059] ​Optionally, in another possible implementation of the first aspect, when the number of points contained in the current line is at least 3, after extracting the geometric structure features of each point of the current line, the step S2 further comprises:

[0060] When the number of points contained in the current line is less than 3, returning to step S2 to continue traversing the next line.

[0061] Beneficial effects: The scheme of the present application first acquires contour line data and wind field data, then traverses all lines in the contour line data, when the number of points contained in the current line is at least 3, extracts the geometric structure features of each point of the current line, then extracts 4 wind field component fields within a preset range of each point in the current line, and calculates the wind direction angle and the wind field type of the component field, and finally constructs a feature matrix of the line according to the geometric structure features of each point of the current line, the wind direction angle and the wind field type. Thus, the present application has the following three beneficial effects: 1. The accuracy of vector coordinates guarantees the mathematical rigor of curvature calculation and convex-concave point detection, avoiding problems such as inaccurate feature positioning caused by pixel graph line boundary blur and scaling drift, and further improving the rejection rate of pseudo-groove-ridge lines through direction features (tangent direction angle, global direction consistency); 2. The wind field component field only needs to extract 4 grid point data around each contour point, without identifying a pixel graph representing a global wind field, thereby improving the recognition efficiency of the machine learning model; 3. The feature matrix of the line is greatly reduced in volume compared with the pixel graph, so the data storage occupancy, transmission time consumption, and recognition efficiency of the machine learning model can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0063] Figure 1 is a flowchart of a vector graph feature extraction method for groove-ridge line recognition provided by an embodiment of the present application;

[0064] Figure 2 is a schematic diagram of a preset range of 4 wind field grid points around a contour point provided by an embodiment of the present application;

[0065] Figure 3 is a flowchart of groove-ridge line recognition provided by an embodiment of the present application;

[0066] Figure 4 is a weather map provided by an embodiment of the present application. DETAILED DESCRIPTION

[0067] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known methods, devices, circuits, and

[0068] It will be understood that the terms "comprises" and / or "comprising," when used in this specification, include the presence of one or more features, integers, steps, operations, elements, and / or components described in the specification, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0069] It will be understood that the term "and / or," when used in the specification and in the following claims, indicates one and / or any combination of the listed terms in the specification and / or claims.

[0070] As used in this specification and in the claims, the terms "if" and "when" can be interpreted to mean "upon determination" or "in response to a determination" or "upon detection" or "in response to a detection" depending on the context. Similarly, the phrase "if determined" or "if detected [the described condition or event]" can be interpreted to mean "upon determination" or "in response to a determination" or "upon detection" or "in response to a detection" depending on the context.

[0071] In addition, the terms "first," "second," "third," etc. in the description of the specification and in the claims are used for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances such that the descriptive terms "first," "second," "third," etc. are to be interpreted, by the reader, as "one or more," "one or more," "one or more," etc. respectively and the terms "first," "second," "third," etc. are not necessarily used in a sequential or chronological order.

[0072] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application. Thus, the appearances of the phrases "in one embodiment" or "in an embodiment" or "in some embodiments" or "in other embodiments" or "in still other embodiments," etc. in various places throughout this specification are not necessarily all referring to the same embodiment, unless otherwise specifically so stated. Furthermore, the term "comprising" or "containing" or "including" or "having" or "with" as used in this specification are specifically intended to be construed as "including but not limited to" unless otherwise specifically indicated otherwise.

[0073] A vector graph feature extraction method for ridge line identification provided by the present application is described in detail below with reference to the accompanying drawings.

[0074] Figure 1 A flowchart of a vector graph feature extraction method for ridge line identification provided by an embodiment of the present application is shown.

[0075] As Figure 1 shown, the vector graph feature extraction method for ridge line identification includes the following steps:

[0076] S1, obtaining contour data and wind field data;

[0077] Specifically, the structure of the contour data in the above step S1 is contour={cloc,ctag}, cloc is a line list, containing cn lines, and the line c∈{c1,c2,...c cn}, c is a longitude and latitude coordinate matrix with a dimension of [cnc,2], where cnc represents the number of points that make up the cth line, and each point contains longitude and latitude data; ctag is a tag sequence, containing cn potential height data; the structure of the wind field data in S1 is wind={wlon,wlat,u,v,w}, where wlon is an equidistant longitude coordinate sequence with a length of wnc; wlat is an equidistant latitude coordinate sequence with a length of wnr; u is a horizontal meridional component field, which is a matrix with a dimension of [wnr,wnc]; v is a horizontal zonal component field, which is a matrix with a dimension of [wnr,wnc]; w is a vertical component field, which is a matrix with a dimension of [wnr,wnc].

[0078] S2, traversing all lines in the contour data, and when the number of points contained in the current line is at least 3, extracting the geometric structure features of each point of the current line;

[0079] Further, in the present application, the current line includes a set of two-dimensional coordinate points c=[p1,p2,...,p cnc ], where p i =(x i ,y i ) represents the position coordinates of the points on the current line, x i is the longitude, and y i is the latitude, and the extraction of the geometric structure features of the current line in the above step S2 includes:

[0080] S21, calculating the first and second derivatives of the points on the current line by difference method;

[0081] S22, calculating the normalized curvature of each point based on the first and second derivatives;

[0082] S23, detecting concave-convex points based on local maximum curvature according to normalized curvature;

[0083] S24, calculating tangent direction angle and global direction consistency of current line.

[0084] It should be noted that the points on the current line include internal points and boundary points.

[0085] Further, in the embodiment of the present application, the above step S21 is specifically:

[0086] The first derivative of the internal point is:

[0087]

[0088] The first derivative of the boundary point is:

[0089] The first point dx1=x2-x1, dy1=y2-y1

[0090] The last point dx cnc =x cnc -x cnc-1 , dy cnc =y cnc -y cnc-1

[0091] The second derivative of the internal point is:

[0092]

[0093] The second derivative of the boundary point is:

[0094] The first point ddx1=dx2-dx1, ddy1=dy2-dy1

[0095] The last point ddx cnc =dx cnc -dx cnc-1 , ddy cnc =dy cnc -dy cnc-1 .

[0096] Further, in the embodiment of the present application, the above step S22 is specifically:

[0097] The curvature of each point is calculated:

[0098]

[0099] Wherein ε=10 -6 is a parameter to prevent division by zero error;

[0100] The curvature of each point is normalized to obtain the normalized curvature of each point:

[0101]

[0102] Further, in the embodiments of the present application, the step S23 is specifically:

[0103] Performing convex point marking:

[0104]

[0105] Performing concave point marking:

[0106]

[0107] Performing boundary point processing:

[0108] cvex1 = ccave1 = cvex cnc = ccave cnc = 0.

[0109] Further, in the embodiments of the present application, the step S24 is specifically:

[0110] Tangent direction angle:

[0111] θ i = arctan2(dy i , dx i )i∈{1,...cnc}

[0112] Where -π < θ i ≤ π, arctan2 is a complete mapping function from coordinates to angles, which realizes the full quadrant coverage and boundary processing ability through two parameters;

[0113] Global direction consistency:

[0114]

[0115] Where cnc j represents the number of points of the jth line, dy i,j is the first order derivative of the latitude of the ith point of the jth line, dx i,j is the first order derivative of the longitude of the ith point of the jth line, θ global is the average of the tangent direction angles of all points in the line list cloc;

[0116] Δθ global,i = min(|θ i - θ global |, 2π - |θ i - θ global |)i∈{1,...cnc}

[0117] Where, Δθglobal,i the deviation between the tangent direction angle of the i-th point of the current line and θ global .

[0118] Further, in the embodiment of the present application, when the number of points contained in the current line is at least 3, after extracting the geometric structure features of each point of the current line, the step S2 further includes:

[0119] When the number of points contained in the current line is less than 3, returning to step S2 to continue traversing the next line.

[0120] S3, extracting four wind field component fields in a preset range of each point in the current line, and calculating the wind direction angle and the wind field type of the component field;

[0121] Further, in the embodiment of the present application, the step S3 of extracting four wind field component fields in a preset range of each point in the current line, and calculating the wind direction angle and the wind field type of the component field, includes:

[0122] In the current line, extracting the wind field component fields of four grid points in a preset range around each point:

[0123] u i =[u i,1 ,u i,2 ,u i,3 ,u i,4 ], v i =[v i,1 ,v i,2 ,v i,3 ,v i,4 ], w i =[w i,1 ,w i,2 ,w i,3 ,w i,4 ]

[0124] Calculating the wind direction angle of each point:

[0125]

[0126] wherein

[0127] Calculating the wind field type of each point:

[0128]

[0129] As a possible implementation, the wind field data corresponding to the current traversed contour line c, i.e. the wind field component fields around the points of the line, are obtained. For each pi, the component fields of the four grid points around it are taken, which is shown as follows: Figure 2 ui = [u i,1 ,u i,2 ,u i,3 ,u i,4 ], v i = [v i,1 ,v i,2 ,v i,3 ,v i,4 ], w i = [w i,1 ,w i,2 ,w i,3 ,w i,4 ].

[0130] S4, constructing a feature matrix of the line according to the geometric structural features of each point of the current line, the wind direction angle and the wind field type.

[0131] Further, in the embodiments of the present application, the step S4 of constructing the feature matrix of the line according to the geometric structural features of each point of the current line, the wind direction angle and the wind field type comprises:

[0132] constructing a feature vector of each point:

[0133]

[0134]

[0135] constructing the feature matrix of the line according to the feature vector of each point:

[0136]

[0137] This application provides a vector feature extraction method for ridge line identification. First, contour line data and wind field data are acquired. Then, all lines in the contour line data are traversed. When the current line contains at least 3 points, the geometric structure features of each point in the current line are extracted. Then, 4 wind field components within a preset range of each point in the current line are extracted, and the wind direction angle and wind field type of the component fields are calculated. Finally, the feature matrix of the line is constructed based on the geometric structure features, wind direction angle, and wind field type of each point in the current line. Therefore, this application has the following three beneficial effects: 1. The accuracy of vector coordinates ensures the mathematical rigor of curvature calculation and convex / concave point detection, avoiding problems such as blurred line boundaries in pixel images and inaccurate feature localization caused by scaling drift. In addition, directional features (tangent direction angle, global direction consistency) further improve the rejection rate of pseudo-groove ridge lines; 2. The wind field component field only needs to extract the data of the four surrounding grid points of each contour point, without the need to identify the pixel image that presents the global wind field, thereby improving the recognition efficiency of the machine learning model; 3. The feature matrix of the line is significantly reduced in size compared to the pixel image, thus improving the technical effect of data storage, transmission time, and recognition efficiency of the machine learning model.

[0138] The following specific embodiment illustrates a vector feature extraction method for groove and ridge line recognition provided in this application.

[0139] like Figure 3 As shown, taking the identification of trough and ridge lines at a height of 500 hPa as an example, the input data are diamond14 (contour lines), diamond11 (wind field uv component field), and diamond4 (wind field w component field) format data issued by the National Meteorological Administration. The specific formats are shown in Tables 1-3 respectively.

[0140] Table 1

[0141]

[0142] Table 1 Note: Height is fixed at 0. Table 2

[0143]

[0144]

[0145] Table 3

[0146]

[0147] The above vector property data is first read using meteorological operational software (such as MiCAPS4) and then visualized. Figure 4Then the three types of groove lines, ridge lines, and non-groove ridge lines (such as saddle-shaped fields, flat contours, etc.) are manually positioned in the weather map, and then a plurality of rectangular frames (latitude and longitude ranges) and labels (three classification types) corresponding to each frame are obtained.

[0148] According to Figure 4 The latitude and longitude range of the framed weather map sub-image is used to screen the corresponding contour and wind field uvw component field. Here, it is noted that each contour point needs to take the wind field grid point data of the surrounding area, so in order to ensure that the contour points at the boundary can also take the wind field data, the wind field data needs to be screened based on the frame range and then expanded by one grid point to the periphery. In this way, the contour data structure contour and the wind field data structure wind of each frame can be obtained according to the data format of diamond 14, 11, and 4.

[0149] Based on the contour and wind of each frame, a vector graph feature extraction method for trough-ridge line identification provided by the present application is used to obtain the corresponding F = {F1, F2,... F cn} for a sample. In this way, according to the time range of the meteorological data 2015-2019, 08:00 and 20:00 (Beijing time) every day, a total of 6000 samples are extracted, including 2400 trough line samples, 2400 ridge line samples, and 1200 non-groove ridge line samples. Then, according to the experience parameters of small data set division, 70% training set and 30% verification set are divided, and finally the training set is obtained. 1680 trough line samples, 1680 ridge line samples, and 840 non-groove ridge line samples; the verification set is 720 trough line samples, 720 ridge line samples, and 360 non-groove ridge line samples.

[0150] Next, in order to compare the recognition effect of the data set and the pixel property sub-image data set of the present application, based on the aforementioned sub-image framed in the weather map, 6000 pixel samples are also generated, and the training set and the verification set are divided in the same way.

[0151] Then, in terms of the selection of the machine learning model, although the data set of the present application is not uniform in terms of sample dimension, it is more suitable for using a model such as Vision-LSTM for classification and recognition, but considering the comparability with the pixel sub-image data set, a CNN classification model mobilenetv3_small_075 in the timm project is still selected. This model is a lightweight variant of MobileNetV3-Small, which can significantly reduce the parameter quantity and calculation cost, and is natively adapted to small data sets and small size multi-channel data input. Therefore, in order to meet the input requirements of the CNN model, the dimension of each sample of the data set of the present application needs to be expanded to a fixed value and the number of channels is 1, wherein max(cn) represents the maximum number of lines of all samples, and max(cnc) represents the maximum number of line points of all samples, j and max(cnc) represents the maximum number of line points of all samples, After the above processing, the verification evaluation indexes of the data set and the pixel sub-image data set of the application are shown in Table 4.

[0152] Table 4

[0153]

[0154] Note: The above three classification (groove line, ridge line, non-groove ridge line) evaluation index calculation rules: (1) Except for the accuracy (global index, classification independent calculation is meaningless), the precision, recall rate and F1 score are calculated according to the classification; (2) The arithmetic average of the precision, recall rate and F1 score of the three classifications is taken as the data in the table to ensure the equal weight of small sample classes.

[0155] As can be seen from Table 4, the vector graph feature data set is significantly higher than the pixel sub-image data set in the accuracy, precision, recall rate and F1 score four core indexes, with an increase of about 5 percentage points. This improvement shows that the vector graph feature not only extracts more comprehensive (high recall rate reflects less missed recognition) and accurate (high precision reflects less recognition error) feature information in groove ridge line recognition, but also to a certain extent overcomes the main defects of information redundancy, distortion and ambiguity of the original pixel graph, thereby providing a higher quality data set basis for efficient recognition of groove ridge line.

[0156] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0157] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A vector feature extraction method for groove and ridge line recognition, characterized in that, include: S1. Obtain contour data and wind field data; S2. Traverse all lines in the contour data. When the current line contains at least 3 points, extract the geometric structure features of each point in the current line. S3. Extract the four wind field components within a preset range for each point in the current line, and calculate the wind direction angle and wind field type of the component fields. S4. Construct the feature matrix of the line based on the geometric structure features, wind direction angle, and wind field type of each point on the current line.

2. The method as described in claim 1, characterized in that, The structure of the contour data in S1 is contour = {cloc, ctag}, where cloc is a list of lines containing cn root lines, and line c ∈ {c1, c2, ..., ctag}. cn }, c is a latitude and longitude coordinate matrix of dimension [cnc,2], where cnc represents the number of points that make up the c-th line, and each point contains two data points: longitude and latitude; ctag is a tag sequence containing cn geopotential height data; the structure of the wind field data in S1 is wind={wlon,wlat,u,v,w}, where wlon is an arithmetic sequence of longitude coordinates with length wnc; wlat is an arithmetic sequence of latitude coordinates with length wrn; u is the horizontal longitude component field, which is a matrix of dimension [wnr,wnc]; v is the horizontal latitudinal component field, which is a matrix of dimension [wnr,wnc]; w is the vertical component field, which is a matrix of dimension [wnr,wnc].

3. The method as described in claim 2, characterized in that, The current line includes a set of two-dimensional coordinate points c = [p1, p2, ..., p...]. cnc ], where p i =(x i ,y i () represents the position coordinates of a point on the current line, x i For longitude, y i For latitude, the extraction of geometric structural features of the current line in S2 includes: Calculate the first and second derivatives of the points on the current line using the finite difference method; The normalized curvature at each point is calculated based on the first and second derivatives; Concave and convex point detection based on local maximum curvature according to standardized curvature; Calculate the tangent direction angle and the global direction consistency of the current line.

4. The method as described in claim 3, characterized in that, The points on the current line include internal points and boundary points. The calculation of the first and second derivatives of the points on the current line using the finite difference method is as follows: The first derivative of an interior point is: The first derivative at the boundary point is: The first point dx1=x2-x1, dy1=y2-y1 Final point dx cnc = x cnc - x cnc-1 , dy cnc = y cnc - y cnc-1 The second derivative of an interior point is: The second derivative at the boundary point is: First point ddx1=dx2-dx1, ddy1=dy2-dy1 Final point ddx cnc = dx cnc -dx cnc-1 , ddy cnc = dy cnc -dy cnc-1 .

5. The method as described in claim 4, characterized in that, The calculation of the normalized curvature at each point based on the first and second derivatives includes: Calculate the curvature at each point: Where ε = 10 -6 Parameters to prevent division by zero errors; The curvature at each point is standardized to obtain the standardized curvature at each point:

6. The method as described in claim 5, characterized in that, The method of detecting concave and convex points based on local maximum curvature according to standardized curvature includes: Perform convex dot marking: Mark the indentation: Boundary point processing: cvex1=ccave1=cvex cnc =ccave cnc =0。 7. The method as described in claim 4, characterized in that, The calculation of the tangent direction angle and the global direction consistency of the current line are as follows: Tangent direction angle: θ i =arctan2(of i ,dx i )i∈{1,…cnc} Where -π < θ i ≤π, arctan2 is a complete mapping function from coordinates to angles, which achieves full quadrant coverage and boundary handling capability in (-π,π] through two parameters; Global directional consistency: CNC j dy represents the number of points on the j-th line. i,j Let dx be the first derivative of the latitude of the i-th point of the j-th line. i,j Let θ be the first derivative of the longitude of the i-th point on the j-th line. global The mean of the tangent direction angles of all points in the line list cloc; Dth global,i =min(|θ i -θ global |,2π-|θ i -θ global |)i∈{1,...cnc} Where, Δθ global,i The tangent direction angle and θ at the i-th point of the current line global The deviation between them.

8. The method as described in claim 2, characterized in that, In step S3, four wind field components are extracted from a preset range for each point on the current line, and the wind direction angle and wind field type of the component fields are calculated, including: Within the current line, extract the wind field component fields of 4 grid points within a preset range surrounding each point: u i =[u i,1 ,u i,2 ,u i,3 ,u i,4 ],v i =[v i,1 ,v i,2 ,v i,3 ,v i,4 ],w i =[w i,1 ,w i,2 ,w i,3 ,w i,4 ] Calculate the wind direction angle at each point: in Calculate the wind field type for each point:

9. The method according to any one of claims 1-8, characterized in that, In step S4, a feature matrix of the line is constructed based on the geometric structure features, wind direction angle, and wind field type of each point on the current line, including: Construct the feature vector for each point: Construct the feature matrix of the line based on the feature vector of each point:

10. The method as described in claim 1, characterized in that, In step S2, after traversing all lines in the contour data and extracting the geometric structure features of each point of the current line when the current line contains at least 3 points, the process further includes: If the current line contains fewer than 3 points, return to step S2 to continue traversing the next line.

Images