A target type recognition method based on track motion feature analysis
By cleaning and extracting features from radar track data, and combining K-means clustering and gradient boosting classifiers, the problems of accuracy and efficiency in radar target identification and classification are solved, and efficient target type identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINESE PEOPLES LIBERATION ARMY UNIT 93209
- Filing Date
- 2024-05-08
- Publication Date
- 2026-05-29
AI Technical Summary
The accuracy and efficiency of radar target identification and classification in existing technologies need to be improved.
By cleaning radar track data, K-means clustering algorithm is used for cluster analysis to mine the data distribution of position, altitude and velocity features, construct feature vectors, and use gradient boosting classifier for model training and classification.
It enables accurate and rapid classification of radar target trajectory data, improving the accuracy and efficiency of identification.
Smart Images
Figure CN118503752B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of radar data processing technology, specifically relating to a target type identification method based on track motion feature analysis. Background Technology
[0002] Radar tracks include aerial and maritime target tracks. Track data mainly includes information such as longitude, latitude, altitude, and speed. Extracted features include two-dimensional geographic coordinates (longitude and latitude), and three dimensions (altitude and speed). Currently, the accuracy and efficiency of radar target identification and classification need to be improved. Summary of the Invention
[0003] (a) Technical problems to be solved
[0004] The technical problem to be solved by this invention is how to provide a target type identification method based on track motion feature analysis to solve the problem of radar target identification and classification.
[0005] (II) Technical Solution
[0006] To address the aforementioned technical problems, this invention proposes a target type identification method based on trajectory motion feature analysis, which includes the following steps:
[0007] S1. Clean the track data, remove outliers, noise and missing data, and build a training set;
[0008] S2. By using the K-means clustering algorithm, the latitude and longitude coordinates, altitude, and velocity motion features of the training set track data are clustered to obtain the set of position, altitude, and velocity cluster center points;
[0009] S3. By mining the data distribution of position, altitude, and speed features in the flight track data, construct a data distribution histogram and extract position, altitude, and speed feature vectors;
[0010] S4. Use a gradient boosting classifier, with the feature vector as input and the category of the track data as the label, to train the model and obtain a trained classification model.
[0011] S5. Use the trained classification model to classify the target trajectory data and obtain the classification results.
[0012] (III) Beneficial Effects
[0013] This invention proposes a target type identification method based on track motion feature analysis. Through in-depth analysis of radar target track data, this invention proposes a novel classification approach. This method demonstrates extremely high accuracy and efficiency in practice, and can accurately and quickly classify target track data, making it highly practical. Attached Figure Description
[0014] Figure 1 This is a flowchart of the target type identification method based on track motion feature analysis of the present invention;
[0015] Figure 2 Here is a flowchart for location feature extraction;
[0016] Figure 3 Here is a flowchart for high-resolution feature extraction;
[0017] Figure 4 This is a flowchart of the velocity feature extraction process. Detailed Implementation
[0018] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
[0019] To address the problem of radar target identification and classification, this invention provides a target type identification method based on track feature analysis. This is a track-level radar target classification method. By extracting target motion features from track data, a feature vector based on the target motion features is constructed. Then, a gradient boosting decision tree classifier is used for model training to achieve rapid classification of radar targets.
[0020] This invention proposes a novel classification approach through in-depth analysis of radar target trajectory data. This method demonstrates extremely high accuracy and efficiency in practice, enabling accurate and rapid classification of target trajectory data, and has significant practical value.
[0021] This invention provides a target type identification system based on trajectory feature analysis, comprising: a data preprocessing module, a cluster analysis module, a feature extraction module, a model training module, and a classification module;
[0022] The data preprocessing module is used to clean the data, remove outliers, noise and missing data, and build a training set;
[0023] The clustering analysis module is used to cluster latitude and longitude coordinates, altitude, and velocity motion features of the training set track data using the K-means clustering algorithm to extract sample features;
[0024] The feature extraction module is used to construct a data distribution histogram and extract feature vectors by mining the data distribution of location features, height features, and speed features.
[0025] The model training module is used to train a model on feature vectors using a gradient boosting classifier.
[0026] The classification module is used to classify target trajectory data and obtain classification results.
[0027] The present invention provides a method for identifying target types based on the analysis of track motion characteristics, including the following steps:
[0028] S1. Clean the track data, remove outliers, noise, and missing data, and construct a training set;
[0029] S2. By using the K-means clustering algorithm, cluster the longitude and latitude coordinate points, altitude, and speed motion characteristics of the track data in the training set to obtain a set of clustering center points for position, altitude, and speed;
[0030] S3. By mining the data distribution of the position characteristics, altitude characteristics, and speed characteristics of the track data, construct a data distribution histogram, and extract the position, altitude, and speed feature vectors;
[0031] S4. Use a gradient boosting classifier, take the feature vectors as input, and the category of the track data as the label for model training to obtain a trained classification model;
[0032] S5. Use the trained classification model to classify the target track data to obtain a classification result.
[0033] Example 1:
[0034] (I) Preprocessing of target track data
[0035] Mainly read the radar target track data after marking, and obtain a training set after preprocessing.
[0036] Specifically, read the radar target track data after marking, store the category identifier, track ID, time, longitude, latitude, altitude, and speed information in the database, and at the same time clean the data, remove outliers, noise, and missing data, perform outlier elimination, and construct a training set.
[0037] Suppose that the current training set includes N types of track data, and each type of track contains M n tracks, where 0 < n ≤ N, then each track point data can be expressed as
[0038] {n,m,T i ,Lon i ,Lat i ,H i ,T i},0 < i ≤ K nm
[0039] 0 < m ≤ M n
[0040] The above formula represents the track data with category identifier n and track ID m, where the time of the i-th track point data is T iLongitude is Lon i Latitude is Lat i The height is H i The speed is V i K nm The number of trackpoints for each track, K, represents the number of trackpoints for different tracks. nm They could be different values.
[0041] (II) Cluster Analysis
[0042] The preprocessed training set was subjected to cluster analysis. The K-means clustering algorithm was used to cluster the two-dimensional geographical location, altitude, and speed features of latitude and longitude.
[0043] K-means clustering is a commonly used clustering algorithm for grouping data. This algorithm divides data into a pre-specified number of clusters, where the `n_clusters` parameter represents the number of clusters (or cluster centers) to be created. K-means clustering attempts to group data points into a specified number of clusters based on the distance between them, ensuring that each data point belongs to the cluster represented by its nearest cluster center.
[0044] S21. Connect to the specified database file and query all longitude and latitude data. Use longitude and latitude as two-dimensional spatial coordinates for clustering. The number of cluster centers p can be set according to the distribution of longitude and latitude location data. nc p can be obtained nc Dimensional location cluster center set Where P j =(X j ,Y j ), 0 <j≤p nc It is a two-dimensional vector. Currently, p is set by n_clusters. nc Set it to 50.
[0045] S22. Query all height data, cluster the height data, and set the number of cluster centroids h based on the distribution of the height data. nc h can be obtained nc 3D high-level cluster centroid set Currently, h is set by configuring n_clusters. nc Set it to 10.
[0046] S23. Query all speed data, cluster the speed data, and set the number of cluster centroids (v) based on the distribution of the speed data. nc v can be obtained nc 1D velocity cluster center set Currently, by setting n_clusters, v nc Set it to 5.
[0047] (3) Feature extraction
[0048] Mainly, it is to explore the data distribution of data position features, height features, and speed features, construct a histogram, and normalize it as the feature vector corresponding to the feature.
[0049] S31. Extract the position feature vector
[0050] The steps for position feature extraction are shown in Figure 2 as follows.
[0051] S311. Calculate the difference array between the track and the clustering center point.
[0052] Assume the nth type of track data, 0 < n ≤ N, and the total number of tracks in this type of track data is M n , traverse each track. For the mth track, 0 < m ≤ M n , traverse the position clustering center points, and calculate the sum of the squares of the Euclidean distances between each position clustering center point and the longitude and latitude of the track points. Since the position clustering center point is a two-dimensional vector, P i =(X j , Y j ), 0 < j ≤ 50, the formula is as follows:
[0053]
[0054] Diff_Tr pnm =[Diff1, Diff2,..., Diff 50 )
[0055] In the above formula, K nm is the number of track points of the mth track in the nth type of track data. The number of track points K nm of each track is different. Diff_Tr pnm is a 50-dimensional array. Each element in the array is the sum of the squares of the Euclidean distances between all the track points of this track and 50 position clustering center points, that is, the difference value between the track and the position clustering center point, which can be used as the position feature of this track.
[0056] S312. Construct the position feature difference index array.
[0057] For the 50-dimensional difference value array obtained in S311, calculate the index of the minimum value in the array, as shown in the following formula:
[0058] Argmin_Diff_Tr pnm =argmin[Diff1, Diff2,..., Diff 50 )
[0059] Argmin_Diff_Tr pnm The index that minimizes the Euclidean distance between the 50 location cluster centers and the trajectory.
[0060]
[0061] In the above formula, Argmin_Diff pn It is a 50-dimensional array, with array elements defaulting to 0, and the index is Argmin_Diff_Tr. pnm The element is set to 1.
[0062] S313. Constructing a histogram of location features
[0063] Iterate through the other tracks of the current nth type of track data and calculate Argmin_Diff_Tr pnm Index value, Argmin_Diff pn Array corresponding index element Add 1.
[0064]
[0065] This constructs a 50-dimensional array of longitude and latitude location feature histograms, Argmin_Diff. pn , which represents the number of times the location cluster center point and the Euclidean distance closest point of that type of track occur.
[0066] Argmin_Diff pn =[Arg1,Arg2,…,Arg 50 ].
[0067] For each type of track data, the difference index value between the cluster center and all tracks is calculated, ultimately forming the Argmin_Diff. pn Array, Argmin_Diff pn The array represents the number of times the cluster center of this type of target location and the closest point in Euclidean distance of this type of track occur.
[0068] Argmin_Diff pn Normalization yields [A1,A2,…,A] 50 ].
[0069] S314. Constructing the location feature vector
[0070] Iterate through all N types of track data, repeating S311-S313, and calculate all Argmin_Diff values. p1 Argmin_Diff p2 ,…,Argmin_Diff pNThe values form an N×50 matrix, which is normalized to obtain a position feature vector of N×50
[0071] S32. Extract the height feature vector
[0072] See the position feature extraction steps in Figure 3 as shown
[0073] S321. Calculate the difference array between the track and the clustering center point
[0074] Assume the nth type of track data, 0 < n ≤ N, and the total number of tracks in this type of track data is M n , traverse each track. For the mth track, 0 < m ≤ M n , traverse the height clustering center points, and calculate the sum of the squares of the Euclidean distances between each height clustering center point and the height of the track points. The formula is as follows
[0075]
[0076] Diff_Tr hnm =[Diff1,Diff2,…,Diff 10
[0077] In the above formula, K nm is the number of track points of the mth track in the nth type of track data. The number of track points K of each track nm is different. Diff_Tr hnm is a 10-dimensional array. Each element in the array is the sum of the squares of the Euclidean distances between all the track points of this track and 10 height clustering center points, that is, the difference value between the track and the height clustering center point, which can be used as the height feature of this track
[0078] S322. Construct a difference index array
[0079] For the 10-dimensional difference value array obtained in S321, calculate the index of the minimum value in the array, as shown in the following formula
[0080] Argmin_Diff_Tr hnm =argmin[Diff1,Diff2,…,Diff 10
[0081] Argmin_Diff_Tr hnm is used as the index of the minimum Euclidean distance between 10 height clustering center points and this track
[0082]
[0083] In the above formula, Argmin_Diffhn It is a 10-dimensional array, with array elements defaulting to 0, and the index is Argmin_Diff_Tr. hnm The element is set to 1.
[0084] S323. Constructing Feature Histograms
[0085] Iterate through the other tracks of the current nth type of track data and calculate Argmin_Diff_Tr hnm Index value, Argmin_Diff hn Array corresponding index element Add 1.
[0086]
[0087] This constructs a 10-dimensional height feature histogram array, Argmin_Diff. hn , which represents the number of times the high cluster center point and the Euclidean distance of that type of track occur to each other.
[0088] Argmin_Diff hn =[Arg1,Arg2,…,Arg 10 ].
[0089] For each type of track, calculate the difference index between the cluster center and all tracks, ultimately forming the Argmin_Diff. hn Array, Argmin_Diff hn The array represents the number of times the cluster center of this type of target appears with the Euclidean distance closest point of this type of track.
[0090] Argmin_Diff hn Normalization yields [H1,H2,…,H] 10 ].
[0091] S324. Constructing Feature Vectors
[0092] Iterate through all N types of track data, repeating S321-S323, and calculate all Argmin_Diff values. h1 Argmin_Diff h2 ,…,Argmin_Diff hN The values are used to form an N×10 matrix, which is then normalized to obtain an N×10 height eigenvector.
[0093] S33, Extracting the velocity feature vector
[0094] See the steps for velocity feature extraction. Figure 4 As shown.
[0095] S331. Calculate the difference array between the track and the clustering center point.
[0096] Assume the nth class of track data, where 0 < n ≤ N, and the total number of tracks in this class of targets is M n , traverse each track. For the mth track, where 0 < m ≤ M n , traverse the velocity clustering center points, and calculate the sum of the squares of the Euclidean distances between each velocity clustering center point and the velocity of the track point. The formula is as follows:
[0097]
[0098] Diff_Tr vnm = [Diff1, Diff2, …, Diff5]
[0099] In the above formula, K nm is the number of track points of the mth track in the nth class of track data. The number of track points K of each track nm is different. Diff_Tr vnm is a 5-dimensional array. Each element in the array is the sum of the squares of the Euclidean distances between all the track points of this track and the 5 velocity clustering center points, that is, the difference value between the track and the velocity clustering center point, which can be used as the velocity feature of this track.
[0100] S332. Construct the difference index array.
[0101] For the 5-dimensional difference value array obtained in S331, calculate the index of the minimum value in the array, as shown in the following formula.
[0102] Argmin_Diff_Tr vnm = argmin[Diff1, Diff2, …, Diff5]
[0103] Argmin_Diff_Tr vnm is used as the index of the minimum Euclidean distance between the 5 velocity clustering center points and this track.
[0104]
[0105] In the above formula, Argmin_Diff vn is a 5-dimensional array. The elements of the array are default to 0, and the element of the index Argmin_Diff_Tr vnm is set to 1.
[0106] S333. Construct the feature histogram
[0107] Traverse the other tracks of the current nth class of track data, and calculate the index value of Argmin_Diff_Tr vnm Index value, Argmin_Diffvn Array corresponding index element Add 1.
[0108]
[0109] This constructs the 5-dimensional velocity feature histogram array Argmin_Diff. vn , which represents the number of times the velocity cluster center and the Euclidean distance of that type of track occur to each other.
[0110] Argmin_Diff vn = [Arg1,Arg2,…,Arg5].
[0111] For each type of track, calculate the difference index between the cluster center and all tracks, ultimately forming the Argmin_Diff. vn Array, Argmin_Diff vn The array represents the number of times the cluster center of this type of target velocity occurs with the Euclidean distance of this type of track.
[0112] Argmin_Diff vn The normalized result is [V1,V2,…,V5].
[0113] S334. Constructing Feature Vectors
[0114] Iterate through all N types of track data, repeating S331-S333, and calculate all Argmin_Diff values. v1 Argmin_Diff v2 ,…,Argmin_Diff vN The values are used to form an N×5 matrix, which is then normalized to obtain an N×5 velocity eigenvector.
[0115] The calculation steps for position features, height features, and velocity features are basically the same. After concatenating the feature vector matrices, a 50+10+5 dimensional feature vector matrix is formed.
[0116]
[0117] (iv) Model Training
[0118] The model is trained using a gradient boosting classifier. The feature vector matrix obtained from the feature extraction step is used as input, and the category of the target trajectory data is used as the label. The classification model is then trained and saved.
[0119] (V) Classification of Target Track Data
[0120] The classification model generated by model training is used to classify the target track data. The track data is used as input, and the feature vector of the track data is calculated according to the position, altitude, and velocity feature extraction methods in the feature extraction step, resulting in a 65-dimensional feature vector [A1, a2, ..., A...]. 50 ,H1,H2,…,H 10 The feature vectors [V1,V2,…,V5] are used as input to the classification model, and the output is the classification result of the target trajectory.
[0121] This invention proposes a novel classification approach through in-depth analysis of radar target trajectory data. This method demonstrates extremely high accuracy and efficiency in practice, enabling accurate and rapid classification of target trajectory data, and has significant practical value.
[0122] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A target type identification method based on trajectory motion feature analysis, characterized in that, The method includes the following steps: S1. Clean the radar target trajectory data, remove outliers, noise and missing data, and build a training set; S2. By using the K-means clustering algorithm, the latitude and longitude coordinates, altitude, and velocity motion features of the training set track data are clustered to obtain the set of position, altitude, and velocity cluster center points; S3. By mining the data distribution of position, altitude, and velocity features in the flight track data, a data distribution histogram is constructed, and position, altitude, and velocity feature vectors are extracted; for the first... Track-like data, array of position feature histograms The height feature histogram array represents the number of times the location cluster center and the nearest point in Euclidean distance for that type of track occur. The velocity feature histogram array represents the number of times the high-level cluster center and the closest point in Euclidean distance for that type of track occur. This represents the number of times the velocity cluster center and the closest point in Euclidean distance for that type of trajectory occur; S4. Use a gradient boosting classifier, with the feature vector as input and the category of the track data as the label, to train the model and obtain a trained classification model. S5. Use the trained classification model to classify the target trajectory data and obtain the classification results; in, The training set includes N types of track data, and each type of track contains One flight path, among which Then each waypoint data is represented as The above formula represents the category identifier as The track ID is The flight track data, of which the first The time for each trackpoint data is... Longitude is Latitude is The height is The speed is , The number of waypoints for each waypoint; In step S3, the step of extracting the location feature vector includes: S311, Calculate the difference array between the trajectory and the cluster center points; Assume the first Track-like data, The total number of tracks in this category of track data is Traversing each path, for the first a flight path, Iterate through the location cluster centers, calculating the sum of squares of the Euclidean distances (longitude and latitude) between each location cluster center and the track point. Since the location cluster centers are two-dimensional vectors... The formula is as follows: In the above formula, For the first Track data The number of waypoints for each track, and the number of waypoints per track. They are all different. It is a 50-dimensional array. Each element in the array is the sum of the squared Euclidean distances between all track points and the 50 location cluster centers, which is the difference between the track and the location cluster centers, and serves as the location feature of the track. S312. Construct a location feature difference index array; For the 50-dimensional difference value array obtained from S311, calculate the index of the minimum value in the array as follows: The index that has the smallest Euclidean distance to the 50 location cluster centers and the flight path; In the above formula, It is a 50-dimensional array, with array elements starting at index 0 by default. Set the element to 1; S313. Constructing a histogram of location features Traverse the current number Other tracks in similar track data, calculate Index value, Array corresponding index element Add 1; This constructs a 50-dimensional array of longitude and latitude location feature histograms. This represents the number of times the location cluster center point and the closest point in Euclidean distance for that type of track appear; ； For each type of track data, calculate the difference index value between the cluster center and all tracks, ultimately forming... Array, This represents the number of times the cluster center of this type of target location and the closest point in Euclidean distance of this type of track occur; Normalization ; S314. Constructing the location feature vector Iterate through all N types of track data, repeating S311-S313, and calculate all... The value forms a A matrix, normalized to obtain a Location feature vector .
2. The target type identification method based on trajectory motion feature analysis as described in claim 1, characterized in that, Step S1 specifically includes: reading the tagged radar target track data, storing the category identifier, track ID, time, longitude, latitude, altitude, and speed information into the database, cleaning the data to remove outliers, noise, and missing data, and constructing a training set.
3. The target type identification method based on trajectory motion feature analysis as described in claim 1, characterized in that, Step S2 specifically includes: S21. Connect to the specified database file, query all longitude and latitude data, cluster the longitude and latitude as two-dimensional spatial coordinates, and set the number of cluster centers based on the distribution of longitude and latitude location data. ,get Dimensional location cluster center set ,in It is a two-dimensional vector; S22. Query all height data, cluster the height data, and set the number of cluster centroids based on the distribution of the height data. ,get 3D high-level cluster centroid set ; S23. Query all speed data, cluster the speed data, and set the number of cluster centroids based on the distribution of the speed data. ,get 1D velocity cluster center set .
4. The target type identification method based on trajectory motion feature analysis as described in claim 3, characterized in that, By setting n_clusters Set to 50, Set to 10, Set it to 5.
5. The target type identification method based on trajectory motion feature analysis as described in claim 4, characterized in that, In step S3, the step of extracting the height feature vector includes: S321. Calculate the difference array between the trajectory and the cluster center points; Assuming the nth type of track data, The total number of tracks in this category of track data is Traversing each path, for the first a flight path, Traverse the height cluster centers and calculate the sum of squares of the height Euclidean distances between each height cluster center and the waypoint, as shown in the following formula: In the above formula, For the first Track data The number of waypoints for each track, and the number of waypoints per track. They are all different. It is a 10-dimensional array. Each element in the array is the sum of the squared Euclidean distances between all track points and the 10 height cluster centers, which is the difference between the track and the height cluster centers, and serves as the height feature of the track. S322, Construct a difference index array; For the 10-dimensional difference array obtained from S321, calculate the index of the minimum value in the array as follows: The index that has the smallest Euclidean distance to the 10 highly clustered centers of the flight path; In the above formula, It is a 10-dimensional array, with array elements starting at index 0 by default. Set the element to 1; S323. Constructing Feature Histograms Traverse the current number Other tracks in similar track data, calculate Index value, Array corresponding index element Add 1; This constructs a 10-dimensional height feature histogram array. This represents the number of times the highly clustered center point and the closest point in Euclidean distance for that type of track appear; ； For each type of track, calculate the difference index value between the cluster center and all tracks, ultimately forming... Array, The array represents the number of times the cluster center of this type of target appears with the Euclidean distance closest point of this type of track; Normalization ; S324. Constructing Feature Vectors Iterate through all N types of track data, repeating S321-S323, and calculate all... The value forms a A matrix, normalized to obtain a Height feature vector .
6. The target type identification method based on trajectory motion feature analysis as described in claim 5, characterized in that, In step S3, the step of extracting the velocity feature vector includes: S331. Calculate the difference array between the trajectory and the cluster center points; Assume the first Track-like data, The total number of tracks in this category of targets is Traversing each path, for the first a flight path, Traverse the velocity cluster centers and calculate the sum of squares of the Euclidean distances between each velocity cluster center and the track point, as shown in the following formula: In the above formula For the first Track-like data The number of waypoints for each track, and the number of waypoints per track. They are all different. It is a 5-dimensional array. Each element in the array is the sum of the squared Euclidean distances between all track points and the 5 velocity cluster centers, which is the difference between the track and the velocity cluster centers, and serves as the velocity feature of the track. S332. Construct a difference index array; For the 5-dimensional difference value array obtained from S331, calculate the index of the minimum value in the array as follows: The index that minimizes the Euclidean distance between the five velocity cluster centers and the trajectory; In the above formula, It is a 5-dimensional array, with the default element being 0, and the index... Set the element to 1; S333, Constructing Feature Histograms Traverse the current number Other tracks in similar track data, calculate Index value, Array corresponding index element Add 1; This constructs a 5-dimensional velocity feature histogram array. , which represents the number of times the velocity cluster center and the Euclidean distance closest point of that type of track occur; ； For each type of track, calculate the difference index value between the cluster center and all tracks, ultimately forming... Array, The array represents the number of times the cluster center of this type of target velocity occurs with the closest point in Euclidean distance for this type of track; Normalization ; S334. Constructing Feature Vectors Iterate through all N types of track data, repeating S331-S333, and calculate all... The value forms a A matrix, normalized to obtain a velocity eigenvectors ; The eigenvector matrices are concatenated to form a 50+10+5 dimensional eigenvector matrix. .
7. The target type identification method based on trajectory motion feature analysis as described in claim 6, characterized in that, Step S4 specifically includes: training the model using a gradient boosting classifier, using the feature vector matrix obtained in step S3 as input, the category of the target trajectory data as the label, training the classification model and saving it.
8. The target type identification method based on trajectory motion feature analysis as described in claim 7, characterized in that, Step S5 specifically includes: classifying the target track data using a classification model generated by model training: taking the track data as input, calculating the feature vector of the track data according to the position feature, altitude feature, and speed feature extraction methods in step S3, and obtaining a 65-dimensional feature vector. Using this feature vector as input to the classification model, the output is the classification result of the target trajectory.