Maritime abnormal target discrimination method based on frequency analysis and grid adaptive extension

Abstract

The application discloses a marine abnormal target discrimination method based on frequency analysis and grid adaptive expansion. A radar detection area is divided into grids with equal longitude and latitude span, and target frequency data in each grid is counted based on historical AIS track data of each type of target. Then, historical speed and acceleration of different types of targets are counted to obtain target normal activity representation expressed by a Gaussian probability density distribution function. A plurality of random samples are obtained from Dirichlet distribution, and target frequency data statistical results are recalculated in sequence to quantize uncertainty of the target normal activity mode. A cross-subspace support area of an arbitrary grid is formed through an adaptive expansion mechanism based on grid length and uncertainty similarity, and the target activity path coincidence average is calculated based on the uncertainty quantization result of the target type normal activity mode and the cross-subspace support area of the arbitrary grid, and the abnormal target is judged by combining the target speed and other feature coincidence calculation.

Description

Technical Field

[0001] This invention relates to a method for identifying abnormal behavior of ships at sea, belonging to the field of data mining. Background Technology

[0002] Maritime targets are characterized by their large number, variety, and complexity. When non-Chinese surface ships or civilian armed vessels illegally operate in China's near-shore waters, they often employ radio silence or false AIS messages, and utilize numerous civilian vessels within waterways and fishing areas as cover, significantly increasing the difficulty of identifying abnormal targets. Identifying abnormal maritime targets is an effective means of improving the automation and intelligence levels of maritime safety supervision. It can promptly detect potential safety hazards, ensure safe navigation, monitor maritime accidents and illegal activities, optimize traffic efficiency, and guarantee the safety and economy of maritime navigation.

[0003] Gridded target activity pattern statistics are a common method for anomaly detection. They involve statistical analysis of the spatiotemporal range and motion characteristics of targets within a local detection area. Based on gridded methods, they use position probability, Bayesian and other analysis methods to calculate the statistical patterns of historical target trajectories in each grid, such as frequency, speed distribution, maneuverability, and length distribution, according to target type and ship length. This supports the detection of whether a ship's position or trajectory deviates from its usual route and the discovery of targets with abnormal motion behavior. It has been widely used in the detection of abnormal ship trajectories. However, directly using gridded methods for anomaly target discrimination suffers from a serious uneven distribution of target activity frequency data within each grid, resulting in poor anomaly discrimination performance and a high false alarm rate. Summary of the Invention

[0004] In large-scale radar detection areas, existing grid-based anomaly detection methods do not fully consider the geometrical spatial independence of ship behavior, resulting in a high false alarm rate. This invention proposes a maritime anomaly target discrimination method based on frequency analysis and adaptive grid expansion. It relies on historical AIS (Automatic Identification System) trajectory data of maritime targets to statistically analyze target activity patterns. The target trajectory data received by radar is used as the data input for real-time anomaly assessment. The method studies an adaptive expansion mechanism based on grid length and uncertainty similarity to construct a cross-subspace support region for arbitrary grids, supporting the calculation of target activity feature consistency. Combined with target speed, heading and other information, a higher quality target activity confidence score is finally obtained, realizing the detection of anomaly targets.

[0005] The method for identifying maritime anomaly targets based on frequency analysis and adaptive grid expansion provided by this invention includes the following steps:

[0006] 1) Divide the radar detection area into grids with equal latitude and longitude spans, and perform statistical analysis on the frequency data of targets within the grids based on historical AIS trajectory data of various types of targets;

[0007] 2) Statistical analysis of historical velocity, acceleration, etc. of different types of targets is performed to obtain a characterization of normal target activity represented by a Gaussian probability density distribution function;

[0008] 3) By obtaining multiple sets of random samples from the Dirichlet distribution and recalculating the statistical results of the target frequency data in sequence, the uncertainty of the normal activity pattern of this type of target can be quantified;

[0009] 4) Research an adaptive expansion mechanism based on grid length and uncertainty similarity to construct cross-subspace support regions for arbitrary grids;

[0010] 5) Based on the quantification results of the uncertainty of the normal activity mode of the target type and the cross subspace support region of arbitrary grid, calculate the mean value of the target activity path conformity, and combine it with the conformity calculation of features such as target speed to complete the judgment of abnormal targets.

[0011] This invention solves the problem of poor anomaly detection and high false alarm rate caused by the geometrical spatial independence of ship behavior in the gridded conformity calculation method for large-area radar detection areas, thus enabling better detection of abnormal ship behavior. Attached Figure Description

[0012] Figure 1 This is a flowchart of the present invention.

[0013] Figure 2 This is a schematic diagram illustrating the construction of the grid intersection subspace support region. Detailed Implementation

[0014] The present invention will be further explained and described below with reference to the accompanying drawings and embodiments.

[0015] Schematic diagram of the present invention as follows Figure 1 As shown, the preferred embodiment process includes:

[0016] 1) Divide the radar detection area into grids with equal latitude and longitude spans, and statistically analyze the target frequency data within each grid based on historical AIS trajectory data for various target types. The preferred implementation process includes:

[0017] First, the radar detection area is divided into grids with equal latitude and longitude spans, represented as:

[0018] W = {grid 1,1 ,grid 1,2 ,...,grid i,j ,...} (1)

[0019] The grid's attributes include: historical ship density, historical traffic density, current ship list, and frequency data for various target types.

[0020]

[0021] In the formula, TargetDesnsity i,j TarfficDesnsity i,j This is a vector containing the activity patterns of various types of ships. k represents different target types, ch represents channel layering information, and i and j represent grid numbers, corresponding to the horizontal and vertical coordinates respectively. LEN itv This represents the target frequency data within the grid, categorized by ship size (Large, Medium, Small), where itv ∈ {Large, Medium, Small}.

[0022] TargetDesnsity i,j = (Density) i,j CargoDensity i,j (3)

[0023] TarfficDesnsity i,j = (Density) i,j CargoDensity i,j (4)

[0024] In the formula, Density i,j CargoDensity represents the target density and traffic density of all ships within the i-th row and j-th column grid. i,j This represents the target density and traffic density of cargo ships within the grid cell in row i and column j.

[0025] The formula for calculating the target density in grid i within a certain month m is as follows:

[0026]

[0027] In the formula, target_count is the total number of targets. Time is the total time that the s-th target is located in grid i during that month. month_m This is the total time of the month, Area grid_i It is the area of ​​grid i, CrossingTimeSum i It is the sum of the time that all targets are located in grid j in that month.

[0028] The formula for calculating the traffic density in grid i within a certain month m is as follows:

[0029]

[0030] In the formula, It is the total number of times the s-th target is located in grid i in that month, Time month_m This is the total time of the month, Area grid_i CrossingCountSum is the area of ​​grid i. i It represents the total number of times all targets pass through grid j in that month.

[0031] Under each waterway model, target frequency data within the grid is statistically analyzed based on historical AIS trajectory data for each type of target. The histogram of target trajectory frequency data for different types and layers is obtained by calculating the following formula:

[0032]

[0033] In the formula, D represents the total number of grid cells, δ represents the latitude and longitude span, and |.| represents the cardinality of the calculation set.

[0034] 2) Statistical analysis of historical velocities and accelerations of different types of targets yields a characterization of normal target activity expressed by a Gaussian probability density distribution function. The preferred implementation process includes:

[0035] The calculation of the historical velocity and acceleration distribution of various types of targets within the waterway is represented by a Gaussian distribution model as follows:

[0036]

[0037] Where k represents different types of targets, u v and Let u represent the target's historical mean and covariance, respectively. Δv and Let represent the target's historical acceleration mean and covariance, respectively.

[0038] At this time, through This represents the 95% confidence interval for the target's speed and acceleration. For speed compliance calculations of various types of targets within the waterway, it involves judging the target's current speed, acceleration, and confidence interval. The relationship is such that if the speed value exceeds the confidence interval, it is judged as a suspicious target.

[0039]

[0040] In the formula, Z α / 2 σ represents the confidence coefficient. m Let N represent the standard deviation and N represent the total number of samples.

[0041] 3) By obtaining multiple sets of random samples from the Dirichlet distribution and recalculating the statistical results of the target frequency data in sequence, the uncertainty of the normal activity pattern of this type of target is quantified. The preferred implementation process includes:

[0042] By obtaining multiple sets of random samples from the Dirichlet distribution and recalculating the statistical results of the target frequency data in sequence:

[0043]

[0044]

[0045] In the formula, [·] num This indicates the generation of num sets of two-dimensional matrices AIS_Fre_Target. k Random samples of uniform size. * indicates dot product operation. This represents the recalculated target trajectory frequency data, which is achieved by multiplying the data in AIS_Fre_Target with each random sample in Dir(x,num) (a total of num samples) to form a three-dimensional data cube with i rows, j columns, and num layers, thereby quantifying the uncertainty of the normal activity mode of this type of target.

[0046] 4) Research an adaptive expansion mechanism based on grid length and uncertainty similarity to construct cross-subspace support regions for arbitrary grids. The preferred implementation process includes:

[0047] For the three-dimensional data cube in formula (11) For any layer of the grid p, an adaptive expansion mechanism based on the similarity of grid length and uncertainty is used to construct the corresponding intersecting subspace support region Ω. p ,like Figure 2 As shown. First, a cross structure with four directional arms is constructed for each grid. Given any grid p, the expansion at its left end stops when a grid point p1 is found to violate one of the following three rules:

[0048] Rule 1: Diff q value (p,p1)<τ1and Diff q value (p1,p1+(1,0))<τ1

[0049] Rule 2: Diff q length (p,p1)<L1

[0050] Rule 3: Diff q value (p,p1)<τ2,if L2<Diffq length (p,p1)<L1

[0051] In the formula, q∈[1,num] represents the construction of support regions on each layer of the three-dimensional data cube. Diff represents the difference in target trajectory frequency data between grids p and p1. q length (p,p1)=|p-p1| represents the distance between grids p and p1, L1 is the preset maximum adaptive grid expansion distance, and τ1 and τ2 are preset thresholds, satisfying τ2<τ1 and L2<L1. Rule 1 not only restricts the difference in target trajectory frequency data between grids p and p1, but also restricts the difference in frequency data between grid p1 and its previous grid p1+(1,0) in the same direction. When the grid expansion length exceeds L2, a stricter threshold τ2 is used to ensure the grid's extension pattern in the region of similar trajectory frequency data. The right, top, and bottom ends of grid p are constructed in a similar manner. In a preferred embodiment of the invention, the empirical parameters L1, L2, τ1, and τ2 are set to 7, 3, 0.8, and 0.5, respectively. After the cross structure is constructed, the support region Ω of grid p is constructed by merging the horizontal ends of all grids located on the vertical end of p. p .

[0052] 5) Based on the quantification results of the uncertainty of the normal activity mode of the target type and the cross subspace support region of arbitrary grids, calculate the mean value of the target activity path conformity, and combine it with the conformity calculation of features such as target speed to complete the judgment of abnormal targets. The preferred implementation process includes:

[0053] The trajectory of a ship from time T1 to T2 can be represented as a raster sequence:

[0054]

[0055] The anomaly level is assessed by evaluating the ship number of the corresponding type of vessel within each grid in the sequence:

[0056]

[0057]

[0058] in, express If the membership degree is abnormal, it is considered a suspicious target and an abnormal target alarm is triggered when the membership degree value is less than a set threshold.

[0059] The above provides a detailed description of the maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion proposed in this invention, which can be widely applied in the field of maritime anomaly target detection.

Claims

1. A method for identifying maritime anomaly targets based on frequency analysis and adaptive grid expansion, characterized in that: 1) Divide the radar detection area into grids with equal latitude and longitude spans, and perform statistical analysis on the frequency data of targets within the grids based on historical AIS trajectory data of various types of targets; 2) Statistical analysis of the historical velocity and acceleration of different types of targets yields a characterization of normal target activity expressed by a Gaussian probability density distribution function; 3) By obtaining multiple sets of random samples from the Dirichlet distribution and recalculating the statistical results of the target frequency data in sequence, the uncertainty of the normal activity mode of this type of target can be quantified; 4) An adaptive expansion mechanism based on grid length and uncertainty similarity is used to construct cross-subspace support regions for arbitrary grids; 5) Based on the quantification results of the uncertainty of the normal activity mode of the target type and the cross subspace support region of the arbitrary grid, calculate the mean value of the target activity path conformity, and combine it with the target velocity feature conformity calculation to complete the judgment of abnormal targets.

2. The maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion according to claim 1, characterized in that: Step (1) further includes: The radar detection area is divided into grids with equal latitude and longitude spans, represented as: (1) in, grid The attributes include: historical vessel density, historical traffic density, current vessel list, and frequency data for various target types. (2) In the formula, TargetDesnsity i,j ， TarfficDesnsity i,j This is a vector containing the activity patterns of various types of ships. k Indicates different types of targets. ch This indicates channel layering information. i, j Indicates the grid number, corresponding to the horizontal and vertical coordinates respectively. LEN itv This represents the target frequency data within the grid, categorized by ship size (long, medium, small). ; （3） （4） In the formula, Density i,j express i OK j CargoDensity: Target density and traffic density of all ships within the grid. i,j express i OK j Target density and traffic density of cargo ships within the grid; Under each waterway model, target frequency data within the grid is statistically analyzed based on historical AIS trajectory data for each type of target. The histogram of target trajectory frequency data for different types and layers is obtained by calculating the following formula: （7） In the formula, D Indicates the total number of grid cells. δ Indicates the span of latitude and longitude. This indicates the cardinality of the set being computed.

3. The maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion according to claim 2, characterized in that: Step (2) further includes: The calculation of the historical velocity and acceleration distribution of various types of targets within the waterway is represented by a Gaussian distribution model as follows: （8） in k Indicates different types of targets. and Let these represent the target's historical mean and covariance, respectively. and Let these represent the target's historical acceleration mean and covariance, respectively; at this point, through This represents the 95% confidence interval for the target's speed and acceleration. For speed compliance calculations of various types of targets within the waterway, it involves judging the target's current speed, acceleration, and confidence interval. The relationship is such that if the speed value exceeds the confidence interval, it is judged as a suspicious target; （9） In the formula, Z α / 2 Represents the confidence coefficient. σ m Indicates standard deviation, N This represents the total number of samples.

4. The maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion according to claim 3, characterized in that: Step (3) further includes: By obtaining multiple sets of random samples from the Dirichlet distribution and recalculating the statistical results of the target frequency data in sequence: (10) (11) In the formula, Indicates generation num Groups and two-dimensional matrices Random samples of uniform size; This represents the dot product operation. This represents the frequency data of the recalculated target trajectory, which is... The data in the middle are respectively with Each set of random samples is multiplied by a dot product, resulting in a total of num Groups, ultimately forming i OK j List num A three-dimensional data cube of layers is used to quantify the uncertainty of the normal activity pattern of this type of target.

5. The maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion according to claim 4, characterized in that: Step (4) also includes: For the three-dimensional data cube in formula (11) Its arbitrary layer of grid p An adaptive expansion mechanism based on grid length and uncertainty similarity is used to construct the cross subspace support region that forms the corresponding grid. First, construct a cross structure with four directional arms for each grid cell, given any grid cell. p When a grid point is found p 1. The expansion on the left side stops when one of the following three rules is violated: Rule 1: ; Rule 2: ; Rule 3: ; In the formula This means constructing support regions on each layer of the three-dimensional data cube. Represents grid p and p The difference in frequency data of target trajectories between 1 and 2 Represents grid p and p The distance of 1 room L 1 represents the preset maximum distance for adaptive grid expansion. τ 1， τ 2 is a preset threshold, and satisfies τ 2< τ 1， L 2< L 1; Rule 1 restricts the grid p and p The difference in target trajectory frequency data between 1 grid cells restricts the grid cells in the same direction. p 1 and its previous grid p The difference in frequency data between 1 and (1,0); when the grid extension length exceeds L When 2, use the threshold. τ 2. To ensure the extension pattern of the raster in similar regions of trajectory frequency data; raster p The right, top, and bottom ends are constructed in a similar manner.

6. The maritime anomaly target discrimination method based on frequency analysis and grid adaptive expansion according to claim 5, characterized in that: Step (5) further includes: Ship from time T 1 arrive T 2 The trajectory up to this point can be represented as a raster sequence: (12) The anomaly level is assessed by evaluating the ship number of the corresponding type of vessel within each grid in the sequence: (13) (14) in, express If the membership degree is abnormal, it is considered a suspicious target and an abnormal target alarm is triggered when the membership degree value is less than a set threshold.

Images