A method for identifying fishing boat behavior based on multi-step clustering of AIS trajectory data

Abstract

The application discloses a fishing boat behavior mode recognition method based on multi-step clustering of AIS track data, which comprises track data processing, extraction of track feature points, establishment of a similarity distance model to obtain a space-time distance matrix between track points, first-step clustering of the track of the fishing boat by using a Mean Shift clustering algorithm, so that space-time distance similar and continuous track points are gathered together, corresponding to the same fishing behavior, clustering of the average speed of the track section by using a K-means clustering algorithm, so that track sections with similar average speeds are gathered together, and thus all track sections of a fishing boat are classified according to anchoring, navigation and fishing behaviors. The method disclosed by the application reduces the calculation cost and improves the efficiency and universality.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fishing boat behavior pattern recognition, and particularly relates to a fishing boat behavior recognition method based on multi-step clustering of AIS trajectory data. BACKGROUND

[0002] A space-time trajectory is a sequence of records of the position and time of a moving object, as an important space-time object data type, the space-time trajectory has wide application in the fields of traffic flow pattern and characteristic research, resource allocation, sea ice monitoring and the like, and through analysis of various space-time trajectory data, similarity features in the space-time trajectory data can be obtained, and meaningful trajectory patterns can be found.

[0003] Fishing boat trajectory data is a kind of space-time trajectory data, and records the sailing process and corresponding behavior features of a fishing boat. With the wide application of an automatic identification system (AIS) on a fishing boat, it is easier and easier to obtain fishing boat trajectory data, and these fishing boat AIS trajectory data including attributes such as ship position, time, ship speed, heading, turning angle speed are data sources for analyzing the aggregation characteristics of fishing boats, and how to mine valuable information from these massive data has important significance for studying the traffic behavior pattern of fishing boats and analyzing the traffic flow characteristics of fishing boats.

[0004] However, the trajectory data only contains the sailing state data of the fishing boat, and does not contain information such as the fishing method, operation state, fishing intensity and fishing ground change of the fishing boat. However, the acquisition of these fishing information first needs to identify the fishing behavior of the fishing boat from the trajectory of the fishing boat, and then the trajectory data can be analyzed in a deeper level. The fishing boat behavior recognition based on the trajectory has important significance for real-time monitoring of the fishing boat behavior, operation fishing ground, fishing intensity, and mastering the change of fishery resources. In terms of monitoring the fishing behavior of the fishing boat, first, the fishing activity of the ship needs to be identified, and then whether the ship is engaged in illegal fishing activities is identified according to the forbidden fishing area and the position of the ship. In terms of exploring the fishing ground, the hot area of fishing of the fishing boat can be counted by analyzing all the fishing boat trajectories, and then the hot area of fishing can be defined as the fishing ground. In terms of fishery resource management, the fishing times are extracted from the AIS trajectory data abroad, and the cumulative fishing amount is calculated for fishery resource assessment, so as to supervise the uncontrolled fishing behavior. In terms of ensuring the safety of aquatic products, the fishing area can be identified according to the trajectory data to realize the traceability of aquatic products, so as to obtain the source of the aquatic products. The implementation of these functions depends on the fishing boat behavior recognition technology based on the trajectory, therefore, it is of great significance to study the technology of identifying the fishing boat behavior from the trajectory.

[0005] The fishing boat behavior can be generally divided into three categories, namely, stopping behavior, fishing behavior and steaming behavior. The stopping behavior refers to the low-speed non-fishing state of the fishing boat in the state of stopping or netting, which can be further divided into the state of being in the port, the state of being at sea and the state of low-speed netting. In the state of being at sea, the fishing boat is dragged by the anchor chain, the trajectory structure is circular or semicircular with the anchor point as the center, and the speed is basically 0. In the state of low-speed netting, the fishing boat is decelerated for a short time after fishing, the trajectory structure is generally individual trajectory point after the fishing trajectory, and the speed is obviously lower than the normal fishing speed. In the state of being in the port, the trajectory points are concentrated on the edge of the land and highly coincide, and the speed is 0. In the fishing behavior, the fishing boat often performs multiple to-and-fro trawl operations in a certain area. In the same fishing section, the fishing speed and the turning angle of the fishing boat are basically unchanged. Usually, the fishing boat will continuously fish for multiple times, and there is a short netting action between the two fishing states. In addition, the heading of the fishing boat between the adjacent two fishing states is generally obviously different, and the fishing trajectory of the fishing boat may even present a Z-shaped turning structure. The steaming behavior can be divided into three states: the first state is the state of entering and leaving the fishing ground, the second state is the state of transferring between fishing grounds, and the third state is the state of short-term navigation after the netting state. In the state of entering and leaving the fishing ground and the state of transferring between fishing grounds, the trajectory structure of the fishing boat is mostly linear. In the state of short-term navigation, the trajectory structure of the fishing boat is mostly a small amount of trajectory points between two fishing states, and is often accompanied by the trajectory points after the netting stopping state. Compared with stopping and fishing, the speed of the steaming state is the highest. The researchers perform the fishing boat behavior recognition by using the multi-dimensional feature distribution of the fishing boat in the above different states.

[0006] The trajectory segment-based clustering method is applied to the fishing boat behavior identification for the first time in 2018 when Zhang proposed the fishing boat behavior identification algorithm based on multi-step clustering (Algorithm Fishing Behavior Identification, MSC-FBI). The MSC-BFI algorithm adopts a multi-dimensional data weighted summation method to establish a trajectory point distance model, and adopts a density-based clustering method with noise (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) to divide the trajectory sub-segments, and aggregates the trajectory sub-segments through a K-means clustering algorithm, so as to realize the clustering and identification of all trajectory points. The trajectory segment-based fishing boat behavior identification algorithm MSC-FBI has problems of parameter sensitivity and poor general effect. The core algorithm DBSCAN needs two parameters: scanning radius (eps) and minimum number of contained points (minPts), which is parameter sensitive and easy to cause batch misjudgment. SUMMARY

[0007] An object of the present application is to provide a fishing boat behavior identification method based on multi-step clustering of AIS trajectory data, which is more efficient, less parameter sensitive, more robust and has better general effect.

[0008] To solve the above technical problems, the technical scheme adopted by the present application is:

[0009] The fishing boat behavior pattern recognition method based on multi-step clustering of AIS trajectory data provided by the present application comprises the following steps:

[0010] S1. Obtain the fishing boat trajectories of different fishing boats based on the AIS trajectory data, each fishing boat trajectory comprising trajectory points at different times, and the trajectory points comprising ship name, speed, heading, position reporting time, latitude and longitude, etc., and then perform abnormal trajectory processing.

[0011] S2. Extract the trajectory feature points of the fishing boat trajectories according to the minimum description length criterion (Minimum Description Length, MDL);

[0012] S3. Establish a similarity distance model;

[0013] S4. Use the clustering algorithm Mean Shift based on kernel density estimation to perform the first step clustering on the trajectory feature points of all fishing boats, so that similar and continuous trajectory points are clustered together;

[0014] S5. Using K-means clustering algorithm to cluster all the clusters obtained after step S4, so that the trajectory segments with similar average speeds are clustered together, and all the trajectory segments are classified according to anchoring, sailing and fishing behaviors;

[0015] Another object of the present application is to provide a computer readable storage medium having stored thereon a computer program which, when executed by a computer, causes the computer to perform the above method.

[0016] Still another object of the present application is to provide a computing device comprising a memory having stored therein executable code and a processor which, when executing the executable code, implements the above method.

[0017] Compared with the prior art, the present application has the following beneficial effects:

[0018] The present application is based on AIS trajectory data, selects feature trajectory points using the minimum description length criterion, reduces the computational overhead, and improves the model efficiency; a weighted summation method is used to establish a similarity distance model between trajectory points, and the fusion of speed distance, heading distance, time distance and spatial distance is realized; the Mean Shift algorithm is a single parameter algorithm, which reduces the parameter sensitivity and is easy to integrate as a module with other algorithms, improving the versatility. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is the overall implementation framework diagram of the method of the present application.

[0020] Figure 2 is an example diagram of trajectory feature points.

[0021] Figure 3 is a schematic diagram of horizontal distance, vertical distance and angle distance. DETAILED DESCRIPTION

[0022] For a further understanding of the present application, the application will be described in detail with reference to the embodiments. The following embodiments are used to explain the present application, but are not used to limit the scope of the present application.

[0023] With reference to the accompanying drawings, the present application relates to a fishing boat behavior pattern recognition method based on multi-step clustering of AIS trajectory data, comprising the following steps: Figure 1

[0024] S1. Obtain the fishing boat trajectory of different fishing boats based on AIS trajectory data, each fishing boat trajectory includes trajectory points at different times, and the trajectory points include ship name, speed, heading, reporting time, latitude and longitude, etc., and then perform abnormal trajectory processing;

[0025] The abnormal trajectory processing is to delete trajectory data with missing data and repeated trajectory data.​

[0026] S2. Extracting trajectory feature points from the fishing vessel trajectory according to the minimum description length (MDL) criterion;

[0027] For a given line segment, the algorithm selects its feature point based on the MDL computation cost of the point as a feature point being less than the MDL computation cost of not selecting it as a feature point. Feature points are selected by successively checking whether each point meets this condition. Specifically:

[0028] S2.1 traverses all track points of each fishing boat and calculates the feature point calculation cost of each track point according to the minimum description length principle and non-feature point calculation overhead ,if , then the current trajectory point is considered to be a trajectory feature point, and the current trajectory point is added to the feature point set X; otherwise, the current trajectory point is considered not to be a trajectory feature point;

[0029] For any fishing vessel's fishing vessel track , ,in represents the trajectory point at the i-th moment; Non-feature point calculation overhead and feature point calculation overhead They are:

[0030] (1),

[0031] (2),

[0032] The function express and The Euclidean distance between them is calculated based on longitude and latitude;

[0033] Refer to the attached Figure 3 As shown, the horizontal distance , vertical distance and angular distance The calculation formulas are:

[0034] (3),

[0035] in, are two different line segments; and Represents the trajectory The two end points relative to vertical distance; function Indicates taking the minimum value, and respectively represent the distance of two end points of two trajectories in horizontal direction; represent the angle of trajectories ;

[0036] Referring to FIG. 6, the non-feature point calculation overhead and the feature point calculation overhead of trajectories Figure 2 , , are respectively:

[0037] (4),

[0038]

[0039]

[0040]

[0041]

[0042]

[0043]

[0044] (5),

[0045] S3. Establish a similarity distance model, specifically:

[0046] S3.1 The speed distance is to measure the difference in speed between trajectory points in the speed dimension. Assuming that the speeds of trajectory points and are , , , , ; the absolute value of the speed difference is used to measure the speed distance, that is, the calculation formula of the speed distance between two trajectory points is:

[0047] (6),

[0048] The heading distance is to describe the difference in heading between trajectory points in the heading dimension. Assuming that the headings of trajectory points and are , , , , ; since 0° and 360° are the same direction, the heading angle is used to measure the heading distance, that is, the calculation formula of the heading distance between two trajectory points is:​​ is:

[0049] (7),

[0050] wherein represents the absolute value of the heading angle difference between the trajectory points and

[0051] The time distance is used to describe the difference in time between the trajectory points in the time dimension, assuming that the reporting times of the trajectory points and are and , , , The absolute value is used to measure the time distance between the trajectory points, i.e., the time distance between the two trajectory points The calculation formula is:

[0052] (8),

[0053] The spatial distance is used to describe the difference between the trajectory points in the spatial dimension, assuming that the longitude and latitude of the trajectory points and are , , , , The Euclidean distance between the two trajectory points is used to measure the spatial distance, i.e., the calculation formula for the spatial distance between the two trajectory points is:

[0054] (9),

[0055] wherein represents the longitude, represents the latitude;

[0056] S3.2. To achieve the fusion of the speed distance, the heading distance, the time distance, and the spatial distance, a weighted sum

[0057] and is used to establish the similarity distance model between the trajectory points. That is, the similarity distance between the trajectory points and The calculation formula is:

[0058] (10),

[0059] wherein, is a weight matrix, is the weight of the speed distance, is the weight of the heading distance, ​​is the weight of time distance, is the weight of spatial distance;

[0060] S4. Use the kernel density estimation based clustering algorithm Mean Shift to cluster the trajectory feature points of all fishing boats in the first step, so that the similarity distance Close and continuous trajectory points are clustered together;

[0061] Because the similarity distance of trajectory points comprehensively considers the speed, direction, and structural characteristics of the trajectory, trajectory points with similarity distances correspond to the same fishing behavior, thereby achieving segmentation of the trajectory according to behavior; specifically:

[0062] S4.1. Initialize the number of iterations k = 1;

[0063] S4.2 Randomly select a trajectory feature point x in the feature point set X k As cluster C k The cluster center of

[0064] S4.3 Find all the k Similarity distance Trajectory feature points smaller than the threshold h are added to the set M k , we believe that these points belong to cluster C k ,remember k , k Represents the set M k The total number of feature points in the inner trajectory;

[0065] S4.4 Calculate the cluster center x k To set M k The drift mean of each trajectory feature point in :

[0066] (11),

[0067] in, Represents the Gaussian kernel function:

[0068] (12),

[0069] S4.5 Update cluster centers:

[0070] (13),

[0071] S4.6 If there are other clusters and the current cluster C k The cluster center x k With some other cluster C e If the similarity distance of the cluster center is less than the threshold h, the current cluster Ck with cluster C e merge, and update the cluster center as the cluster center of C e ; update k=k+1; if there is no other cluster, directly update k=k+1;

[0072] S4.7 Repeat steps S4.2-S4.6 until the convergence condition is met;

[0073] S5. Use the K-means clustering algorithm to perform secondary clustering on all clusters obtained after step S4, so that trajectory segments with similar average speeds are clustered together, and classify all trajectory segments according to anchoring, sailing and fishing behaviors; specifically:

[0074] S5.1 The set O of all clusters obtained in step S4 is , M, where cluster contains the set of trajectory feature points , , calculate the average speed of all trajectory feature points in each cluster :

[0075] (14),

[0076] where denotes the speed of trajectory feature point ;

[0077] S5.2 Randomly select 3 different clusters from set O as 3 initial cluster centers, corresponding to the classes of anchoring, fishing and sailing behaviors , and ;

[0078] S5.3 Calculate the average speed difference of each cluster in set O to the 3 cluster centers , and divide the cluster into the class corresponding to the cluster center with the smallest value:

[0079] (15), where

[0080] denotes the average speed of or or or ;

[0081] S5.4 Recalculate the cluster center Q of each class :

[0082] (16),

[0083] wherein represents a class number of clusters in represents a class a certain cluster in represents a cluster average speed of all track feature points in the cluster

[0084] S5.5 repeat steps S5.3-S5.4 until the convergence condition is met; compare the average speeds of the three classes finally obtained, from small to large, representing the berthing behavior state, the fishing behavior state and the sailing behavior state respectively;

[0085] The application has been described in detail with reference to the embodiments above, but the content described is only the preferred embodiments of the application and cannot be considered as limiting the scope of the implementation of the application. Any equivalent changes and improvements made in the scope of the application should still belong to the patent coverage of the application.

Claims

1. A fishing vessel behavior recognition method based on multi-step clustering of AIS trajectory data, characterized by The following steps are involved: S1. Obtaining the trajectories of different fishing vessels based on AIS trajectory data, and then performing abnormal trajectory processing on the trajectories of the fishing vessels; The fishing vessel track of each fishing vessel includes track points at different times, including vessel name, speed, heading, reporting time, longitude and latitude; S2. Extracting trajectory feature points of the fishing vessel trajectory according to the minimum description length criterion; S3. Establish a similarity distance model, as follows: S3.1 Speed ​​distance is the difference in speed between trajectory points in the speed dimension. Assume that trajectory point p i and p j The speeds are speed i , speed j , 1≤i≤N, 1≤j≤N, i≠j; the absolute value of the speed difference is used to measure the speed distance, that is, the speed distance between two trajectory points is calculated as: V(i,j)=|speed i -speed j | (6) The heading distance describes the difference in heading between trajectory points in the heading dimension. Assume that the trajectory point p i and p j The headings are degree i ,degree j , 1≤i≤N, 1≤j≤N, i≠j; since heading 0° and 360° are in the same direction, the heading angle is used to measure the heading distance, that is, the heading distance D(i,j) between trajectory points is: Where △degree represents the trajectory point p i and p j The absolute value of the heading angle difference; Time distance describes the temporal differences between trajectory points in the time dimension. Assume that the trajectory point p i and p j The reporting time is time i and time j , 1≤i≤N, 1≤j≤N, i≠j; the absolute value is used to measure the time distance between trajectory points, that is, the time distance T(i,j) between two trajectory points is calculated as: T(i,j)=|time i -time j | (8), Spatial distance describes the difference between trajectory points in the spatial dimension. Assume that trajectory point p i and p j The longitude and latitude are <lon i ,lat i >, <lon j ,lat j >, 1≤i≤N, 1≤j≤N, i≠j; the Euclidean distance between two trajectory points is used to measure the spatial distance, that is, the spatial distance calculation formula between two trajectory points is: Where lon represents longitude and lat represents latitude; S3.

2. In order to achieve the fusion of speed distance, heading distance, time distance and space distance, a weighted summation method is used to establish the similarity distance model between trajectory points, that is, trajectory point p i and p j The similarity distance S(i,j) is calculated as follows: S(i,j)=ω V ×V(i,j)+ω D ×D(i,j)+ω T ×T(i,j)+ω S ×L(i,j) (10), Where ω={ω V ,ω D ,ω T ,ω S } is the weight matrix, ω V is the weight of speed distance, ω D is the weight of the heading distance, ω T is the weight of time distance, ω S is the weight of spatial distance; S4. Use the kernel density estimation-based clustering algorithm Mean Shift to perform the first step of clustering on the trajectory feature points of all fishing vessels, so that the similarity distance S(i, j) is close and continuous trajectory points are clustered together; S5. Use the K-means clustering algorithm to perform secondary clustering on all clusters obtained after processing in step S4, so that trajectory segments with similar average speeds are clustered together, and all trajectory segments are classified according to anchoring, sailing, and fishing behaviors.

2. A fishing vessel behavior recognition method based on multi-step clustering of AIS trajectory data as claimed in claim 1, characterized in that The abnormal trajectory processing in step S1 is to delete the trajectory data with missing data and the repeated trajectory data.

3. A fishing vessel behavior recognition method based on multi-step clustering of AIS trajectory data as claimed in claim 1, characterized in that Step S2 is as follows: traverse all the track points of each fishing boat, calculate the feature point calculation cost MDL of each track point according to the minimum description length principle par and non-feature point calculation overhead MDL nopar If MDL par >MDL nopar , then the current trajectory point is considered to be a trajectory feature point, and the current trajectory point is added to the feature point set X; otherwise, the current trajectory point is considered not to be a trajectory feature point; For any fishing vessel trajectory {p1,p2,...,p i ...,p N }, 1≤i≤N, where p i represents the trajectory point at the i-th moment; p i Non-feature point calculation overhead MDL nopar And feature point calculation overhead MDL par They are: MDL nopar =log2(len(p1p i )) (1) The function len(p1p i ) indicates that p1 and p i The Euclidean distance between them is calculated based on longitude and latitude; Vertical distance d ⊥ and angular distance d θ The calculation formulas are: Among them, L α , L β For two different trajectories, α≠β; and Represents the trajectory L α The two end points are relative to L β The vertical distance; θ represents the trajectory L α , L β Angle.

4. A fishing vessel behavior recognition method based on multi-step clustering of AIS trajectory data as claimed in claim 1, characterized in that Step S4 specifically includes: S4.

1. Initialize the number of iterations k = 1; S4.2 Randomly select a trajectory feature point x in the feature point set X k As cluster C k The cluster center of S4.3 Find all the k Trajectory feature points whose similarity distance S(i,j) is less than the threshold h are added to the set M k , we believe that these points belong to cluster C k , remember x i ∈M k ,i=1,2,...,n k , n k Represents the set M k The total number of feature points in the inner trajectory; S4.4 Calculate the cluster center x k To set M k The drift mean M of each trajectory feature point h (x k ): Among them, K() represents the Gaussian kernel function: S4.5 Update cluster centers: x k =x k +M h (x k ) (13), S4.6 If there are other clusters and the current cluster C k The cluster center x k With some other cluster C e If the similarity distance of the cluster center is less than the threshold h, the current cluster C k With cluster C e Merge and update the cluster center to cluster C e At the same time, update k = k + 1 and repeat steps S4.2-S4.6 until the convergence condition is met; if there are no other clusters, directly update k = k + 1.

5. A fishing vessel behavior recognition method based on multi-step clustering of AIS trajectory data as claimed in claim 1, characterized in that Step S5 specifically includes: S5.1 The set O of all clusters obtained in step S4 is (C1, C2, ..., C m ,...,C M ), 1≤m≤M, where cluster C m The set of trajectory feature points contained in is (u1,u2,...,u e ,...,u E ), 1≤e≤E, calculate the average speed of all trajectory feature points in each cluster where v e Represents the trajectory feature point u e speed; S5.2 Randomly select 3 different clusters C from the set O a ,C b ,C c ∈O, as the three initial cluster centers, corresponding to the class R of the three behavior states of anchoring, fishing and sailing a , R b and R c ; S5.3 Calculate the average speed difference from each cluster in set O to the three cluster centers Divide the cluster into In the class corresponding to the cluster center with the smallest value: in Indicates C a or C b or C c average speed; S5.4 For each class R r ∈(R a ,R b ,R c ) Recalculate the respective cluster centers Q: where |R r | represents class R r The number of clusters, C r Represents class R r A cluster in Represents cluster C r The average speed of all trajectory feature points in ; S5.5 Repeat steps S5.3-S5.4 until the convergence condition is met; compare the average speeds of the three classes finally obtained, which represent the mooring behavior state, fishing behavior state and navigation behavior state from small to large respectively.

6. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 5.

7. A computing device comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method according to any one of claims 1 to 5 is implemented.

Images