Ship gill net catching behavior discrimination method and system based on track spatial-temporal characteristics

Abstract

The invention provides a ship gill net catching behavior discrimination method and system based on track spatial-temporal characteristics, and the method comprises the steps: obtaining ship track data, and carrying out the descending collection according to a fixed interval, and obtaining a target track; target track preliminary screening: filtering out ships in normal sailing to obtain a preliminary screening result; marking gill net fishing targets from the preliminary screening result as training data; constructing a decision tree model, and carrying out training data preprocessing, feature engineering, model training and test improvement; incorporating the decision tree model into an algorithm, and opening algorithm configuration modification parameters; and identifying and storing the track data and the feature data of the gill net target into a database. According to the method, through spatial-temporal feature extraction and behavior pattern recognition of the track, the precision and efficiency of ship behavior monitoring are remarkably improved, gill net fishing behaviors can be recognized more accurately, an efficient and reliable solution is provided for fishery administration departments, fishery resources are better protected, and ecological balance is better maintained.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular, relates to a method and system for distinguishing ship gillnet fishing behavior based on the spatiotemporal characteristics of tracks. Background Art

[0002] In recent years, illegal gillnet fishing has still occurred frequently in coastal and inland waters, causing serious damage to fishery resources. Traditional fishery supervision methods have problems with insufficient coverage and slow response speed when facing vast sea areas and complex navigation environments. Therefore, it is necessary to introduce advanced technical means to conduct intelligent monitoring and early warning of gillnet fishing, and the track behavior of ships is one of the important features for identifying gillnet fishing activities.

[0003] Existing ship track analysis technologies usually rely on traditional data processing methods. These methods mainly perform basic processing on track data, including data smoothing and behavior judgment based on preset rules. The limitation of traditional methods is that they fail to fully combine the temporal and spatial characteristics of the track, which leads to obvious deficiencies in the judgment of gillnet fishing behavior.

[0004] For example, the presence of abnormal behavior can be identified based on fixed standards of parameters such as the ship's speed, heading, and position. However, these traditional methods often have problems with insufficient accuracy and reliability when faced with complex navigation environments, especially when identifying dynamic and subtle fishing behaviors.

[0005] Patent document "A method and device for identifying drifting gillnet operation of fishing vessel" (CN115934785A) discloses the use of feature parameter sets to identify different gillnet operation states through supervised learning, which mainly relies on extracting the track data of fishing vessels and establishing the corresponding relationship between operation and navigation characteristics. However, it lacks flexibility, data labeling is insufficient, and it relies too much on quantitative analysis of track data.

[0006] The wandering behavior identification method disclosed in the patent document "Ship wandering behavior identification method and system based on track spatiotemporal characteristics" (CN117392434A) does not define gillnet fishing behavior and has relatively poor comprehensive identification ability.

[0007] In order to solve these problems, the present invention introduces advanced data analysis technology, including spatiotemporal feature extraction of track and behavior pattern recognition, which can more accurately identify gillnet fishing behavior, significantly improve the accuracy and efficiency of ship behavior monitoring, and provide fishery management departments with an efficient and reliable solution, which helps to better protect fishery resources and maintain ecological balance. Summary of the invention

[0008] In view of the defects in the prior art, the object of the present invention is to provide a method and system for distinguishing ship gillnet fishing behavior based on the spatiotemporal characteristics of the track.

[0009] According to the present invention, a method for distinguishing ship gillnet fishing behavior based on track spatiotemporal characteristics includes:

[0010] Step S1: Obtain ship track data and obtain the target track by sampling at fixed intervals;

[0011] Step S2: Initial screening of target tracks, filtering out ships in normal navigation to obtain initial screening results;

[0012] Step S3: marking gillnet fishing targets from the initial screening results as training data;

[0013] Step S4: Build a decision tree model, perform training data preprocessing, feature engineering, train the model and test the improvements;

[0014] Step S5: Incorporate the decision tree model into the algorithm, open the algorithm configuration to modify parameters, and obtain the gillnet fishing recognition algorithm;

[0015] Step S6: Identify the gillnet target and save the gillnet target track data and feature data into the database.

[0016] Preferably, in step S1, the track data T of the ship is obtained by radar, AIS or Beidou equipment. n}, keep each track data point interval to one point per minute, reduce the sampling time interval Δt = 60s, only calculate the track data of the most recent hour, denoted as N = 60, n∈N, the target track after reduced sampling is T′ = {t′1, t′2, ..., t′ n}.

[0017] The track data is a list stored in chronological order, including the ship's unique identification, heading, speed, longitude and latitude coordinates, length and time.

[0018] The fixed interval reduction is '

[0019] T≤N;

[0020] Among them, t i Represents the track data of the i-th;

[0021] t′ i Represents the track data of the ith one after the drop;

[0022] t′ last Indicates the track data at the last time point after the drop;

[0023] n means there are track data of n time points in total.

[0024] In the step S2, all the track data are preliminarily screened to filter out the target tracks that are obviously normal driving, stationary target tracks, and anchored target tracks.

[0025] The target track that is obviously normal driving is a target track where the number of points of the target track traveling in a straight line at a low speed is greater than 10.

[0026] The filtering is to add a low-speed point condition to the target track after the downsampling. Each target speed is less than 2 knots and the continuous low-speed points are greater than 15. The initial screening result is recorded as T" = {L1, L2, ..., L n}.

[0027] The low-speed point condition is to set the target track T′ after the reduction to { t′1,t′2,...,t′ n} Split into continuous straight line segments L j = {P i ,P i+1 ,…P i+k}, the heading difference ΔL(P i ,P i+k ) is less than the heading difference threshold of 20 degrees set by courseDIffTh, and the number of track points in each straight line segment is greater than 5;

[0028] For each straight line segment L j = {P i ,P i+1 ,…P i+k}Calculate driving distance k∈j; the straight line segment with the maximum travel distance in each target track Driving distance

[0029] Among them, d(P k ,P k+1 ) are two adjacent points P k , P k+1 The distance between

[0030] L represents a straight line segment.

[0031] In step S3, the initial screening results of the target tracks are combined with visualization tools to manually analyze the track shape characteristics, label the targets, and at the same time, the track data that are excluded as gillnet fishing targets are used as counterexample samples.

[0032] Preferably, in step S4, when observing the track of the gillnet fishing target, its track characteristics are analyzed, the data is preprocessed through feature engineering, and features with high discrimination are extracted. A plurality of feature combinations are used as inputs of a decision tree for training, and the feature threshold is calculated by the decision tree. Test data is selected to verify the model, and the tree is optimized through adjustable parameters in the decision tree.

[0033] The features defined in the feature engineering include the number of slow-speed straight-line driving points, the proportion of slow-speed straight-line driving points, the proportion of points in the speed ranges of 0-2, 2-4 and 4-6, the average length, the length variance, the low-speed point heading variance, the low-speed point heading change average, the low-speed point heading change rate variance, the number of round-trip points in the track, the average speed, the low-speed point speed variance, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the low-speed section track, the proportion of the maximum distance of the low-speed section track and the ratio of the maximum distance of the track to the cumulative sailing length.

[0034] The high-discrimination features include the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the variance of the heading change rate of low-speed points, the variance of the azimuth change rate of low-speed points, the variance of the speed of low-speed points, the maximum distance of continuous point tracks in the low-speed section, the ratio of the maximum distance of point tracks to the cumulative navigation length, and the ratio of the maximum distance of point tracks to the number of track points.

[0035] Preferably, the number of points for the low-speed straight-line driving is the number of track points in the target low-speed straight-line driving. n } Traverse each straight line segment L i ={v1,v2,…v m}, track point speed v j Less than 2, j∈m, calculate the number of low-speed points in each straight line segment sum(L i ) are accumulated as the low-speed straight-line driving points sumLowV,

[0036] The low-speed straight-line driving point ratio is the ratio of the track points in the target low-speed straight-line driving to the total track points.

[0037] The variance of the heading change rate at the low speed point is T" = { L1,L2,…,L n} Traverse each track L i ={C1,C2,…,C m}, calculate the heading change value Δθ between two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθ i =360°-Δθi ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', the average heading change rate of each section Variance i The formulas for the total target track heading change rate variance AvgVariance are:

[0038] Where m represents the number of points in each straight line segment;

[0039] n represents the number of straight line segments in each target;

[0040] C m Represents the heading information of the mth track point in the track;

[0041] The heading change value is calculated for the track points of the target traveling at a low speed, with clockwise being positive and counterclockwise being negative, and the average value of the overall heading change is taken.

[0042] The variance of the azimuth change rate of the low-speed point is to directly calculate the azimuth instead of the heading value to make the track shape feature obvious. The track data used is T = {t1, t2, ..., t n}, the heading value is calculated using the azimuth in degrees.

[0043] The speed variance of the low speed point is T" = { L1,L2,…,L n} Calculate the variance of the speed.

[0044] The maximum distance between the continuous points in the low speed section is T”= { L1,L2,…,L n} Calculate the distance between the two farthest points in the track, calculate each low-speed straight track segment separately, and take the maximum value.

[0045] The ratio of the maximum distance of the point track to the cumulative navigation length is the ratio of the distance between the two farthest points in the target track to the target cumulative navigation length.

[0046] The ratio of the maximum distance of the point track to the number of track points is the ratio of the distance between the two farthest points in the target track to the number of track points, and the average distance between every two points of the target is calculated.

[0047] Preferably, in step S5, the geographical information and related configuration data of the fishing area are determined in combination with environmental factors in the actual project, and the recognition logic of the gillnet fishing target is adjusted;

[0048] The training method of the decision tree model belongs to a white box model, and the training logic of the decision tree model for gillnet fishing target analysis is incorporated into the code.

[0049] In step S6, the target track data and feature data identified by the gillnet fishing recognition algorithm are stored in a database, and a warning is issued for the identified gillnet target in the prohibited fishing area.

[0050] According to the present invention, a ship gillnet fishing behavior identification system based on track spatiotemporal characteristics includes:

[0051] Module M1: Obtain ship track data and obtain the target track by sampling at fixed intervals;

[0052] Module M2: Initial screening of target tracks, filtering out ships in normal navigation to obtain initial screening results;

[0053] Module M3: Marking gillnet fishing targets from the initial screening results as training data;

[0054] Module M4: Build a decision tree model, perform training data preprocessing, feature engineering, train the model and test improvements;

[0055] Module M5: Incorporate the decision tree model into the algorithm, open the algorithm configuration to modify the parameters, and obtain the gillnet fishing recognition algorithm;

[0056] Module M6: Identify gillnet targets and save gillnet target track data and feature data into the database.

[0057] Preferably, the module M1 obtains the ship's trajectory data T={t1, t2, ..., t n}, keep each track data point interval to one point per minute, reduce the sampling time interval Δt = 60s, only calculate the track data of the most recent hour, denoted as N = 60, n∈N, the target track after reduced sampling is T′ = {t′1, t′2, ..., t′ n}.

[0058] The track data is a list stored in chronological order, including the ship's unique identification, heading, speed, longitude and latitude coordinates, length and time.

[0059] The fixed interval reduction is '

[0060] T≤N;

[0061] Among them, t i Represents the track data of the i-th;

[0062] t′ i Represents the track data of the ith one after the drop;

[0063] t′ last Indicates the track data at the last time point after the drop;

[0064] n means there are track data of n time points in total.

[0065] In the module M2, all track data are preliminarily screened to filter out tracks of targets that are obviously traveling normally, tracks of stationary targets, and tracks of anchored targets.

[0066] The target track that is obviously normal driving is a target track where the number of points of the target track traveling in a straight line at a low speed is greater than 10.

[0067] The filtering is to add a low-speed point condition to the target track after the downsampling. Each target speed is less than 2 knots and the continuous low-speed points are greater than 15. The initial screening result is recorded as T" = {L1, L2, ..., L n}.

[0068] The low-speed point condition is to set the target track T′ after the reduction to { t′1,t′2,...,t′ n} Split into continuous straight line segments L j = {P i ,P i+1 ,…P i+k}, the heading difference ΔL(P i ,P i+k ) is less than the heading difference threshold of 20 degrees set by courseDIffTh, and the number of track points in each straight line segment is greater than 5;

[0069] For each straight line segment L j = {P i ,P i+1 ,…P i+k}Calculate driving distance k∈j; the straight line segment with the maximum travel distance in each target track Driving distance

[0070] Among them, d(P k ,P k+1 ) are two adjacent points P k , P k+1 The distance between

[0071] L represents a straight line segment.

[0072] In the module M3, the initial screening results of the target tracks are combined with visualization tools to manually analyze the track shape characteristics, label the targets, and at the same time, the track data that are excluded as gillnet fishing targets are used as counterexample samples.

[0073] Preferably, in the module M4, when observing the track of the gillnet fishing target, the track characteristics are analyzed, the data is preprocessed through feature engineering, and features with high discrimination are extracted. Multiple feature combinations are used as inputs of the decision tree for training, the feature threshold is calculated by the decision tree, test data is selected to verify the model, and the tree is optimized through adjustable parameters in the decision tree.

[0074] The features defined in the feature engineering include the number of slow-speed straight-line driving points, the proportion of slow-speed straight-line driving points, the proportion of points in the speed ranges of 0-2, 2-4 and 4-6, the average length, the length variance, the low-speed point heading variance, the low-speed point heading change average, the low-speed point heading change rate variance, the number of round-trip points in the track, the average speed, the low-speed point speed variance, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the low-speed section track, the proportion of the maximum distance of the low-speed section track and the ratio of the maximum distance of the track to the cumulative sailing length.

[0075] The high-discrimination features include the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the variance of the heading change rate of low-speed points, the variance of the azimuth change rate of low-speed points, the variance of the speed of low-speed points, the maximum distance of continuous point tracks in the low-speed section, the ratio of the maximum distance of point tracks to the cumulative navigation length, and the ratio of the maximum distance of point tracks to the number of track points.

[0076] Preferably, the number of points for the low-speed straight-line driving is the number of track points in the target low-speed straight-line driving. n}Traverse each straight line segment L i ={v1,v2,…v m}, track point speed v j Less than 2, j∈m, calculate the number of low-speed points in each straight line segment sum(L i ) are accumulated as the low-speed straight-line driving points sumLowV,

[0077] The low-speed straight-line driving point ratio is the ratio of the track points in the target low-speed straight-line driving to the total track points.

[0078] The variance of the heading change rate at the low speed point is T" = { L1,L2,…,L n} Traverse each track L i ={C1,C2,…,C m}, calculate the heading change value Δθ between two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθi =360°-Δθ i ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', the average heading change rate of each section Variance i The formulas for the total target track heading change rate variance AvgVariance are:

[0079] Where m represents the number of points in each straight line segment;

[0080] n represents the number of straight line segments in each target;

[0081] C m Represents the heading information of the mth track point in the track;

[0082] The heading change value is calculated for the track points of the target traveling at a low speed, with clockwise being positive and counterclockwise being negative, and the average value of the overall heading change is taken.

[0083] The variance of the azimuth change rate of the low-speed point is to directly calculate the azimuth instead of the heading value to make the track shape feature obvious. The track data used is T = {t1, t2, ..., t n}, the heading value is calculated using the azimuth in degrees.

[0084] The speed variance of the low speed point is T" = { L1,L2,…,L n} Calculate the variance of the speed.

[0085] The maximum distance between the continuous points in the low speed section is T”= { L1,L2,…,L n} Calculate the distance between the two farthest points in the track, calculate each low-speed straight track segment separately, and take the maximum value.

[0086] The ratio of the maximum distance of the point track to the cumulative navigation length is the ratio of the distance between the two farthest points in the target track to the target cumulative navigation length.

[0087] The ratio of the maximum distance of the point track to the number of track points is the ratio of the distance between the two farthest points in the target track to the number of track points, and the average distance between every two points of the target is calculated.

[0088] Preferably, the module M5 determines the geographical information and related configuration data of the fishing area in combination with the environmental factors in the actual project, and adjusts the recognition logic of the gillnet fishing target;

[0089] The training method of the decision tree model belongs to a white box model, and the training logic of the decision tree model for gillnet fishing target analysis is incorporated into the code.

[0090] The module M6 stores the target track data and feature data identified by the gillnet fishing recognition algorithm in a database, and issues a warning to the identified gillnet target in the prohibited fishing area.

[0091] Compared with the prior art, the present invention has the following beneficial effects:

[0092] 1. The present invention extracts spatiotemporal features of the track and recognizes behavioral patterns, extracts and classifies features using a decision tree model, dynamically adjusts algorithm parameters and combines necessary conditions in different waters, and can more accurately identify gillnet fishing behavior.

[0093] 2. The present invention defines gillnet fishing behavior and can effectively identify gillnet fishing behavior. The analysis results are stored for subsequent supervision and use, providing an efficient and reliable solution for fishery management departments, which helps to better protect fishery resources and maintain ecological balance.

[0094] 3. The present invention pre-processes the data through data filtering and feature engineering, and adopts steps such as data downsampling, preliminary screening, and manual labeling to extract discriminative features, improve the efficiency of model training, and combine the environmental factors in actual projects to significantly improve the accuracy and efficiency of ship behavior monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0096] Figure 1 This is a flow chart of the method for identifying ship gillnet fishing behavior based on the spatiotemporal characteristics of tracks.

[0097] Figure 2 Schematic diagram of the dynamic anchoring target track.

[0098] Figure 3 Schematic diagram of the architecture of the ship gillnet fishing behavior identification system based on the spatiotemporal characteristics of the track. DETAILED DESCRIPTION

[0099] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0100] The definition of ship gillnet fishing behavior is that in a specific water area, the target ship is traveling at a low speed to set or collect the net, and its track is a thorn-shaped, linear low-speed motion trajectory. The possibility of the ship engaging in gillnet fishing is determined by analyzing the track behavior characteristics of the target ship. The present invention can effectively identify and supervise gillnet fishing behavior and is suitable for supervision scenarios for protecting fishery resources. According to a ship gillnet fishing behavior identification method based on track spatiotemporal characteristics provided by the present invention, Figure 1 For example, including:

[0101] Step S1: Obtain ship track data, track point data and regional location data at fixed intervals.

[0102] The ship’s trajectory data T = {t1, t2, ..., t n} and conduct in-depth analysis of these data to determine whether the target vessel is engaged in gillnet fishing.

[0103] The ship track data includes the ship's unique identification, heading, speed, latitude and longitude coordinates, ship length, and time. The track data of each ship is a list stored in chronological order. In order to improve the difference of calculated features and enhance the distinguishability of features, such as speed variance features and speed change rate features, the point track data of the track is downsampled. The target track after downsampling is T′= { t′1,t′2,...,t′ n} , keep the track point interval of each target at one point per minute, and the time interval of downsampling is Δt=60, unit: seconds. Only the track data of the most recent hour is calculated, that is, N=60, unit: number of points, which can better train the model.

[0104] The fixed interval data reduction formula is as follows:

[0105]

[0106] Among them, t i Represents the track data of the i-th;

[0107] t′ i Represents the track data of the ith one after the drop;

[0108] t′ last Indicates the track data at the last time point after the drop;

[0109] n means there are track data of n time points in total.

[0110] Step S2: Initial screening of targets, filtering out ships in normal voyage.

[0111] In the step S2, all tracks are preliminarily screened to filter out most of the target tracks that are obviously normal driving, stationary target tracks, and anchored target tracks, so as to reduce the amount of calculated data and exclude irrelevant track data.

[0112] In more preferred examples, model training requires positive and negative samples. Since there is not yet a large amount of track data confirmed as gillnet targets, manual labeling is required. When the number of targets is too large, the efficiency of marking will be greatly affected, so the data needs to be filtered. Filtering normal tracks and stationary targets reduces the amount of data that needs to be observed.

[0113] During the fishing process, the gillnet target has a small course change rate and a low speed, and is in slow motion. The basis for the initial screening is that the number of points where the target track travels in a straight line at a low speed is greater than 10. The specific calculation steps are:

[0114] First, filter the target track after the sampling is reduced, and add the low-speed point condition: each target speed is less than 2 knots, and the continuous low-speed points are greater than 15.

[0115] For the filtered downsampled track T′= { t′1,t′2,...,t′ n} , dividing it into continuous track straight line segments L j = {P i ,P i+1 ,…P i+k}, the heading difference ΔL(P i ,P i+k ) must be less than the set heading difference threshold courseDIffTh=20 degrees, and the number of track points contained in each straight line segment must be greater than 5.

[0116] For each straight line segment L j = {P i ,P i+1 ,…P i+k}Calculate the driving distance and require the straight line segment with the maximum driving distance in each target Its driving distance Meters, straight line segment L j The calculation formula for driving distance is:

[0117]

[0118] Among them, d(P k ,P k+1 ) are two adjacent points P k , P k+1 The distance between them, k∈j, unit: meter. The target that meets the above conditions is the track that passes the initial screening, denoted by T”={L1,L2,…,Ln}, where L represents each straight line segment, which is also a necessary condition for subsequent gillnet target judgment.

[0119] Step S3: marking gillnet fishing targets from the initial screening results as training data;

[0120] The steps of initial screening and subsequent data preprocessing are emphasized, and manual analysis of labeled data is used to improve the quality of training samples. In step S3, the gillnet fishing characteristics are analyzed and calculated, and the target tracks after initial screening are combined with visualization tools, and the track shape characteristics are manually analyzed to label the targets as positive samples of subsequent model training data. At the same time, the track data that are excluded as gillnet fishing targets are used as negative samples.

[0121] When processing unlabeled or small amounts of labeled data, the lack of a training data set may affect the accuracy and robustness of the model, and there may be insufficient data labeling. By using downsampling and initial screening to filter out ships that are sailing normally, the quality and efficiency of labeled data can be improved, making the positive and negative samples more balanced.

[0122] Step S4: decision tree modeling, data preprocessing, feature engineering, model training and improvement, and model testing.

[0123] In the step S4, the track characteristics of the target gillnet fishing track are analyzed when observing the track, and the data is preprocessed by feature engineering to extract the discriminative features. In the feature definition process, a total of 17 features are defined: the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the proportion of points in the speed (0-2, 2-4, 4-6) interval, the average length, the length variance, the low-speed point heading variance, the low-speed point heading change average, the low-speed point heading change rate variance, the number of round-trip track points, the average speed, the low-speed point speed variance, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the track in the low-speed section (multiple sections take the maximum value), the proportion of the maximum distance of the track in the low-speed section, and the ratio of the maximum distance of the track to the cumulative sailing length. Through a combination of multiple features as the input of the decision tree, after multiple tests, 8 features with high discriminative power for the gillnet target track are selected as the input of the decision tree for training, and the feature threshold is calculated by the decision tree.

[0124] The highly discriminative features include:

[0125] Slow-speed straight-line driving points: The number of track points of the target in the slow-speed straight-line driving. For track T”= { L1,L2,…,L n}Traverse each straight line segment L i ={v1,v2,…v m}, track point speed v j (j∈m) needs to be less than 2, and the number of low-speed points in each straight line segment sum(Li ) are accumulated as the low-speed straight-line driving points sumLowV, and the specific calculation formula is:

[0126]

[0127] Slow-speed straight-line driving point ratio: The ratio of the target’s track points in slow-speed straight-line driving to the total track points.

[0128] The variance of the heading change rate at the low-speed point is: for the target’s low-speed straight-line track, that is, the initial screening result T”={L1,L2,…,L n}Traverse each track L i ={C1,C2,…,C m}, and then calculate the heading change value Δθ of two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθ i =360°-Δθ i ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', then the average heading change rate of each section is Variance i , the formulas of the total target track heading change rate variance AvgVariance are:

[0129]

[0130] Where m is the number of points in each straight line segment;

[0131] n is the number of straight line segments in each target;

[0132] C m Represents the heading information of the mth track point in the track;

[0133] The heading change value is calculated for the target's low-speed track points, with clockwise being positive and counterclockwise being negative. The average value of the overall heading change is taken, so that the calculated variance value of the "thorn-shaped" track will also be larger.

[0134] The variance of the azimuth change rate at low speed points: The heading value in the track data is smoothed. The calculated value of some targets is smaller than the track shape of the target that can be seen intuitively. Therefore, the azimuth is directly calculated instead of the heading value to make the track shape feature obvious. Compared with the variance feature of the heading change rate at low speed points, the difference is that the track data used is T = {t1, t2, ..., t n} Total track, heading value is calculated in degrees using azimuth.

[0135] Over-reliance on quantitative analysis of track data may lead to neglect of potential shape features and other nonlinear relationships in the track data. Therefore, we not only focus on quantitative features, but also combine comprehensive feature analysis of multiple features such as track shape changes to improve the comprehensive recognition ability of the model, especially the discrimination of “thorny” tracks.

[0136] Speed ​​variance at low speed point: The low speed track point T” of the target { L1,L2,…,L n} Calculate the variance of the speed.

[0137] Maximum distance of continuous point track in low speed section: The low speed track section T” = { L1,L2,…,L n} Calculate the maximum distance of its activity range, that is, the distance between the two farthest points in the track. Calculate each low-speed straight track segment separately. If there are multiple low-speed track segments, take the maximum value.

[0138] Ratio of maximum distance between points and cumulative sailing length: Ratio of the distance between the two farthest points in the target track and the cumulative sailing length of the target. This feature is used to filter dynamic anchor targets. Figure 2 For example, the characteristic of a dynamic anchoring target is that the ship remains anchored in place but its track will shift in all directions, creating the illusion of short-distance, low-speed movement.

[0139] Ratio of maximum track distance to track points: The ratio of the distance between the two farthest points in the target track to the number of track points. The average distance between every two points of the target is calculated. This feature is used to filter anchored and small-range drifting targets.

[0140] Test data is selected to verify the model, and the tree is optimized through the adjustable parameters in the decision tree.

[0141] Step S5: Incorporate the decision tree model into the algorithm for secondary improvement, open the algorithm configuration parameters, and modify the parameters to adapt to the environment.

[0142] Compared with the traditional method, which relies more on fixed feature parameters and models and is difficult to cope with complex or changeable environments, a dynamic decision tree model is adopted, and environmental changes are taken into account in model training, so that feature thresholds and parameters can be adjusted in real time to adapt to different water environments and dynamic conditions, and it is more flexible to adapt to practical applications and enhance flexibility. In step S5, due to limited manpower, the amount of data of the marked gillnet fishing target is small, and only relying on the decision tree for identification may lead to problems such as overfitting, unstable tree structure, and poor model generalization ability. Therefore, it is necessary to determine the geographical information and related configuration data of the fishing area in combination with the environmental factors in the actual project, and adjust the recognition logic of the gillnet fishing target. For example, if there are more floating objects in some waters, the characteristic thresholds of the variance of the course change rate and the speed variance of the low-speed point are set for the floating objects, and if there are more wild animals on the water in some waters, they will be detected as targets by radar, and the ship length screening conditions can be set for the wild animals.

[0143] The training method of decision tree belongs to white box model, which can clearly show its training logic. Therefore, it is decided to incorporate the training logic of decision tree model for gillnet fishing target analysis into the code so that it can be flexibly adjusted according to the project environment.

[0144] Step S6: Identify and save the gillnet target track data and feature data into a database.

[0145] In step S6, the target track data and feature data identified by the gillnet fishing recognition algorithm used in the actual project are stored in a database for subsequent optimization of the gillnet fishing target recognition model.

[0146] The present invention selects typical gillnet fishing target track feature data and trains the model through machine learning algorithms. Based on the training model and the necessary conditions of different waters, the gillnet fishing behavior is effectively identified and an alarm is issued for the identified gillnet target in the prohibited fishing area. The analysis results will be stored for subsequent supervision, such as warning and arresting illegal fishing behaviors in the prohibited fishing area.

[0147] The present invention also provides a ship gillnet fishing behavior discrimination system based on track temporal and spatial characteristics. The ship gillnet fishing behavior discrimination system based on track temporal and spatial characteristics can be realized by executing the process steps of the ship gillnet fishing behavior discrimination method based on track temporal and spatial characteristics, that is, those skilled in the art can understand the ship gillnet fishing behavior discrimination method based on track temporal and spatial characteristics as a preferred implementation of the ship gillnet fishing behavior discrimination system based on track temporal and spatial characteristics.

[0148] Those skilled in the art know that, in addition to realizing the system and its various devices, modules, and units provided by the present invention in a purely computer-readable program code, it is entirely possible to realize the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a hardware component, and the devices, modules, and units included therein for realizing various functions can also be regarded as structures within the hardware component; the devices, modules, and units for realizing various functions can also be regarded as both software modules for realizing the method and structures within the hardware component.

[0149] According to a ship gillnet fishing behavior identification system based on track spatiotemporal characteristics provided by the present invention, Figure 3 For example, including:

[0150] Module M1: Obtain ship track data and obtain the target track by sampling at fixed intervals;

[0151] Module M2: Initial screening of target tracks, filtering out ships in normal navigation to obtain initial screening results;

[0152] Module M3: Marking gillnet fishing targets from the initial screening results as training data;

[0153] Module M4: Build a decision tree model, perform training data preprocessing, feature engineering, train the model and test improvements;

[0154] Module M5: Incorporate the decision tree model into the algorithm, open the algorithm configuration to modify the parameters, and obtain the gillnet fishing recognition algorithm;

[0155] Module M6: Identify gillnet targets and save gillnet target track data and feature data into the database.

[0156] In more preferred embodiments, the module M1 obtains the ship's trajectory data T={t1, t2, ..., t n}, keep each track data point interval to one point per minute, reduce the sampling time interval Δt = 60s, only calculate the track data of the most recent hour, denoted as N = 60, n∈N, the target track after reduced sampling is T′ = {t′1, t′2, ..., t′ n}.

[0157] The track data is a list stored in chronological order, including the ship's unique identification, heading, speed, longitude and latitude coordinates, length and time.

[0158] The fixed interval reduction is T′≤N;

[0159] Among them, t i Represents the track data of the i-th;

[0160] t′ i Represents the track data of the ith one after the drop;

[0161] t′ last Indicates the track data at the last time point after the drop;

[0162] n means there are track data of n time points in total.

[0163] In the module M2, all track data are preliminarily screened to filter out tracks of targets that are obviously traveling normally, tracks of stationary targets, and tracks of anchored targets.

[0164] The target track that is obviously normal driving is a target track where the number of points of the target track traveling in a straight line at a low speed is greater than 10.

[0165] The filtering is to add a low-speed point condition to the target track after the downsampling. Each target speed is less than 2 knots and the continuous low-speed points are greater than 15. The initial screening result is recorded as T" = {L1, L2, ..., L n}.

[0166] The low-speed point condition is to set the target track after the reduction T′={t′1, t′2, ..., t′ n} is divided into continuous straight line segments L j = {P i ,P i+1 ,…P i+k}, the heading difference ΔL(P i ,P i+k ) is less than the heading difference threshold of 20 degrees set by courseDIffTh, and the number of track points in each straight line segment is greater than 5;

[0167] For each straight line segment L j = {P i ,P i+1 ,…P i+k}Calculate driving distance k∈j; the straight line segment with the maximum travel distance in each target track Driving distance

[0168] Among them, d(P k ,P k+1 ) are two adjacent points P k , P k+1 The distance between

[0169] L represents a straight line segment.

[0170] In the module M3, the initial screening results of the target tracks are combined with visualization tools to manually analyze the track shape characteristics, label the targets, and at the same time, the track data that are excluded as gillnet fishing targets are used as counterexample samples.

[0171] In more preferred examples, the module M4 analyzes the track characteristics when observing the track of the gillnet fishing target, preprocesses the data through feature engineering, extracts features with high discrimination, trains the decision tree through multiple feature combinations as inputs, calculates the feature threshold by the decision tree, selects test data to verify the model, and optimizes the tree through adjustable parameters in the decision tree.

[0172] The features defined in the feature engineering include the number of slow-speed straight-line driving points, the proportion of slow-speed straight-line driving points, the proportion of points in the speed ranges of 0-2, 2-4 and 4-6, the average length, the length variance, the low-speed point heading variance, the low-speed point heading change average, the low-speed point heading change rate variance, the number of round-trip points in the track, the average speed, the low-speed point speed variance, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the low-speed section track, the proportion of the maximum distance of the low-speed section track and the ratio of the maximum distance of the track to the cumulative sailing length.

[0173] The high-discrimination features include the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the variance of the heading change rate of low-speed points, the variance of the azimuth change rate of low-speed points, the variance of the speed of low-speed points, the maximum distance of continuous point tracks in the low-speed section, the ratio of the maximum distance of point tracks to the cumulative navigation length, and the ratio of the maximum distance of point tracks to the number of track points.

[0174] In more preferred embodiments, the low-speed straight-line driving points are the track points of the target low-speed straight-line driving. { L1,L2,…,L n} Traverse each straight line segment L i ={v1,v2,…v m}, track point speed v j Less than 2, j∈m, calculate the number of low-speed points in each straight line segment sum(L i ) are accumulated as the low-speed straight-line driving points sumLowV,

[0175]

[0176] The low-speed straight-line driving point ratio is the ratio of the track points in the target low-speed straight-line driving to the total track points.

[0177] The variance of the heading change rate at the low speed point is T" = { L1,L2,…,L n}Traverse each track L i ={C1,C2,…,C m}, calculate the heading change value Δθ between two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθ i =360°-Δθ i ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', the average heading change rate of each section Variance i The formulas for the total target track heading change rate variance AvgVariance are:

[0178] Where m represents the number of points in each straight line segment;

[0179] n represents the number of straight line segments in each target;

[0180] C m Represents the heading information of the mth track point in the track;

[0181] The heading change value is calculated for the track points of the target traveling at a low speed, with clockwise being positive and counterclockwise being negative, and the average value of the overall heading change is taken.

[0182] The variance of the azimuth change rate of the low-speed point is to directly calculate the azimuth instead of the heading value to make the track shape feature obvious. The track data used is T = {t1, t2, ..., t n}, the heading value is calculated using the azimuth in degrees.

[0183] The speed variance of the low speed point is T" = { L1,L2,…,L n} Calculate the variance of the speed.

[0184] The maximum distance between the continuous points in the low speed section is T”= { L1,L2,…,L n} Calculate the distance between the two farthest points in the track, calculate each low-speed straight track segment separately, and take the maximum value.

[0185] The ratio of the maximum distance of the point track to the cumulative navigation length is the ratio of the distance between the two farthest points in the target track to the target cumulative navigation length.

[0186] The ratio of the maximum distance of the point track to the number of track points is the ratio of the distance between the two farthest points in the target track to the number of track points, and the average distance between every two points of the target is calculated.

[0187] In more preferred examples, the module M5 determines the geographical information and related configuration data of the fishing area in combination with the environmental factors in the actual project, and adjusts the recognition logic of the gillnet fishing target;

[0188] The training method of the decision tree model belongs to a white box model, and the training logic of the decision tree model for gillnet fishing target analysis is incorporated into the code.

[0189] The module M6 stores the target track data and feature data identified by the gillnet fishing recognition algorithm in a database, and issues a warning to the identified gillnet target in the prohibited fishing area.

[0190] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for distinguishing ship gillnet fishing behavior based on the spatiotemporal characteristics of track, characterized in that: include: Step S1: Obtain ship track data and obtain the target track by sampling at fixed intervals; Step S2: Initial screening of target tracks, filtering out ships in normal navigation to obtain initial screening results; Step S3: marking gillnet fishing targets from the initial screening results as training data; Step S4: Build a decision tree model, perform training data preprocessing, feature engineering, train the model and test the improvements; Step S5: Incorporate the decision tree model into the algorithm, open the algorithm configuration to modify parameters, and obtain the gillnet fishing recognition algorithm; Step S6: Identify the gillnet target and save the gillnet target track data and feature data into the database.

2. The method for distinguishing ship gillnet fishing behavior based on track spatiotemporal characteristics according to claim 1, characterized in that: In step S1, the track data T of the ship is obtained by radar, AIS or Beidou equipment. n }, keep each track data point interval to one point per minute, reduce the sampling time interval Δt = 60s, only calculate the track data of the most recent hour, denoted as N = 60, n∈N, the target track after reduced sampling is T′ = {t′1, t′2, ..., t′ n }; The track data is a list stored in chronological order, including the ship's unique identification, heading, speed, latitude and longitude coordinates, length and time; The fixed interval reduction is Among them, t i Represents the track data of the i-th; t′ i Represents the track data of the ith one after the drop; t′ last Indicates the track data at the last time point after the drop; n means there are track data at n time points; In step S2, all track data are preliminarily screened to filter out target tracks that are obviously in normal driving, stationary target tracks, and anchored target tracks; The target track that is obviously in normal driving is a target track where the number of points of the target track traveling in a straight line at a low speed is greater than 10; The filtering is to add a low-speed point condition to the target track after the downsampling. Each target speed is less than 2 knots and the continuous low-speed points are greater than 15. The initial screening result is recorded as T" = {L1, L2, ..., L n }; The low-speed point condition is to set the target track after the reduction T′={t′1, t′2, ..., t′ n } is divided into continuous straight line segments L j = {P i , P i+1 , …P i+k }, the heading difference ΔL(P i , P i+k ) is less than the heading difference threshold of 20 degrees set by courseDIffTh, and the number of track points in each straight line segment is greater than 5; For each straight line segment L j = {P i , P i+1 , …P i+k }Calculate driving distance The straight line segment with the maximum travel distance in each target track Driving distance Among them, d(P k , P k+1 ) are two adjacent points P k , P k+1 The distance between L represents a straight line segment; In step S3, the initial screening results of the target tracks are combined with visualization tools to manually analyze the track shape characteristics, label the targets, and at the same time, the track data that are excluded as gillnet fishing targets are used as counterexample samples.

3. The method for distinguishing ship gillnet fishing behavior based on track spatiotemporal characteristics according to claim 1, characterized in that: In step S4, when observing the track of the gillnet fishing target, its track characteristics are analyzed. Preprocess the data through feature engineering to extract features with high discriminative power, train the decision tree with multiple feature combinations as input, calculate the feature threshold by the decision tree, select test data to verify the model, and optimize the tree through the adjustable parameters in the decision tree; The features defined in the feature engineering include the number of slow-speed straight-line driving points, the proportion of slow-speed straight-line driving points, the proportion of points in the speed ranges of 0-2, 2-4 and 4-6, the average length, the variance of the length, the variance of the heading at slow-speed points, the average heading change at slow-speed points, the variance of the heading change rate at slow-speed points, the number of round-trip points on the track, the average speed, the variance of the speed at slow-speed points, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the track in the slow-speed section, the proportion of the maximum distance of the track in the slow-speed section, and the ratio of the maximum distance of the track to the cumulative sailing length; The high-discrimination features include the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the variance of the heading change rate of low-speed points, the variance of the azimuth change rate of low-speed points, the variance of the speed of low-speed points, the maximum distance of continuous point tracks in the low-speed section, the ratio of the maximum distance of point tracks to the cumulative navigation length, and the ratio of the maximum distance of point tracks to the number of track points.

4. The method for distinguishing ship gillnet fishing behavior based on track spatiotemporal characteristics according to claim 3, characterized in that: The number of low-speed straight-line driving points is the number of track points in the target low-speed straight-line driving. n }Traverse each straight line segment L i = {v1, v2, ... v m }, track point speed v j Less than 2, j∈m, calculate the number of low-speed points in each straight line segment sum(L i ) are accumulated as the low-speed straight-line driving points sumLowV, The low-speed straight-line driving point ratio is the ratio of the target low-speed straight-line driving track points to the total track points; The variance of the heading change rate at the low speed point is T″={L1, L2, ..., L n }Traverse each track L i ={C1, C2, ..., C m }, calculate the heading change value Δθ between two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθ i =360°-Δθ i ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', the average heading change rate of each section The formulas for the variance of the rate of change Variancei and the variance of the total target track heading rate of change AvgVariance are respectively Where m represents the number of points in each straight line segment; n represents the number of straight line segments in each target; C m Represents the heading information of the mth track point in the track; Calculate the heading change value for the target's low-speed track points, with clockwise being positive and counterclockwise being negative, and take the average value of the overall heading change; The variance of the azimuth change rate of the low-speed point is to directly calculate the azimuth instead of the heading value to make the track shape feature obvious. The track data used is T = {t1, t2, ..., t n }, the heading value is calculated using the azimuth in degrees; The speed variance of the low-speed point is T" = {L1, L2, ..., L n }Calculate the variance of the speed; The maximum distance of the continuous points in the low-speed section is T" = {L1, L2, ..., L n }Calculate the distance between the two farthest points in the track, calculate each low-speed straight track segment separately, and take the maximum value; The ratio of the maximum distance of the point track to the cumulative navigation length is the ratio of the distance between the two farthest points in the target track to the target cumulative navigation length; The ratio of the maximum distance of the point track to the number of track points is the ratio of the distance between the two farthest points in the target track to the number of track points, and the average distance between every two points of the target is calculated.

5. The method for distinguishing ship gillnet fishing behavior based on track spatiotemporal characteristics according to claim 1, characterized in that: In step S5, the geographical information and related configuration data of the fishing area are determined in combination with the environmental factors in the actual project, and the recognition logic of the gillnet fishing target is adjusted; The training method of the decision tree model belongs to a white box model, and the training logic of the decision tree model for gillnet fishing target analysis is incorporated into the code; In step S6, the target track data and feature data identified by the gillnet fishing recognition algorithm are stored in a database, and a warning is issued for the identified gillnet target in the prohibited fishing area.

6. A ship gillnet fishing behavior identification system based on track temporal and spatial characteristics, characterized in that: include: Module M1: Obtain ship track data and obtain the target track by sampling at fixed intervals; Module M2: Initial screening of target tracks, filtering out ships in normal navigation to obtain initial screening results; Module M3: Marking gillnet fishing targets from the initial screening results as training data; Module M4: Build a decision tree model, perform training data preprocessing, feature engineering, train the model and test improvements; Module M5: Incorporate the decision tree model into the algorithm, open the algorithm configuration to modify the parameters, and obtain the gillnet fishing recognition algorithm; Module M6: Identify gillnet targets and save gillnet target track data and feature data into the database.

7. The ship gillnet fishing behavior identification system based on track spatiotemporal characteristics according to claim 6 is characterized in that: The module M1 obtains the ship's trajectory data T={t1, t2, ..., t n }, keep each track data point interval to one point per minute, reduce the sampling time interval Δt = 60s, only calculate the track data of the most recent hour, denoted as N = 60, n∈N, the target track after reduced sampling is T′ = {t′1, t′2, ..., t′ n }; The track data is a list stored in chronological order, including the ship's unique identification, heading, speed, latitude and longitude coordinates, length and time; The fixed interval reduction is Among them, t i Represents the track data of the i-th; t′ i Represents the track data of the ith one after the drop; t′ last Indicates the track data at the last time point after the drop; n means there are track data at n time points; In the module M2, all track data are preliminarily screened to filter out tracks of targets that are obviously traveling normally, tracks of stationary targets, and tracks of anchored targets; The target track that is obviously in normal driving is a target track where the number of points of the target track traveling in a straight line at a low speed is greater than 10; The filtering is to add a low-speed point condition to the target track after the downsampling. Each target speed is less than 2 knots and the continuous low-speed points are greater than 15. The initial screening result is recorded as T" = {L1, L2, ..., L n }; The low-speed point condition is to set the target track after the reduction T′={t′1, t′2, ..., t′ n } is divided into continuous straight line segments L j = {P i , P i+1 , …P i+k }, the heading difference ΔL(P i , P i+k ) is less than the heading difference threshold of 20 degrees set by courseDIffTh, and the number of track points in each straight line segment is greater than 5; For each straight line segment L j = {P i , P i+1 , …P i+k }Calculate driving distance The straight line segment with the maximum travel distance in each target track Driving distance Among them, d(P k , P k+1 ) are two adjacent points P k , P k+1 The distance between L represents a straight line segment; In the module M3, the initial screening results of the target tracks are combined with visualization tools to manually analyze the track shape characteristics, label the targets, and at the same time, the track data that are excluded as gillnet fishing targets are used as counterexample samples.

8. The ship gillnet fishing behavior identification system based on track spatiotemporal characteristics according to claim 6 is characterized in that: The module M4 analyzes the track characteristics of the target gillnet fishing track when observing the track. Preprocess the data through feature engineering to extract features with high discriminative power, train the decision tree with multiple feature combinations as input, calculate the feature threshold by the decision tree, select test data to verify the model, and optimize the tree through the adjustable parameters in the decision tree; The features defined in the feature engineering include the number of slow-speed straight-line driving points, the proportion of slow-speed straight-line driving points, the proportion of points in the speed ranges of 0-2, 2-4 and 4-6, the average length, the variance of the length, the variance of the heading at slow-speed points, the average heading change at slow-speed points, the variance of the heading change rate at slow-speed points, the number of round-trip points on the track, the average speed, the variance of the speed at slow-speed points, the ratio of the maximum distance of the track to the number of track points, the maximum distance of the track, the maximum distance of the track in the slow-speed section, the proportion of the maximum distance of the track in the slow-speed section, and the ratio of the maximum distance of the track to the cumulative sailing length; The high-discrimination features include the number of low-speed straight-line driving points, the proportion of low-speed straight-line driving points, the variance of the heading change rate of low-speed points, the variance of the azimuth change rate of low-speed points, the variance of the speed of low-speed points, the maximum distance of continuous point tracks in the low-speed section, the ratio of the maximum distance of point tracks to the cumulative navigation length, and the ratio of the maximum distance of point tracks to the number of track points.

9. The ship gillnet fishing behavior identification system based on track spatiotemporal characteristics according to claim 8, characterized in that: The number of low-speed straight-line driving points is the number of track points in the target low-speed straight-line driving. n }Traverse each straight line segment L i = {v1, v2, ... v m }, track point speed v j Less than 2, j∈m, calculate the number of low-speed points in each straight line segment sum(L i ) are accumulated as the low-speed straight-line driving points sumLowV, The low-speed straight-line driving point ratio is the ratio of the target low-speed straight-line driving track points to the total track points; The variance of the heading change rate at the low speed point is T″={L1, L2, ..., L n }Traverse each track L i ={C1, C2, ..., C m }, calculate the heading change value Δθ between two adjacent points i '=C i -C i-1 , if Δθ i '>180°, then Δθ i =360°-Δθ i ', if Δθ i '<-180°, then Δθ i =-360°-Δθ i ', the average heading change rate of each section Variance i The formulas for the total target track heading change rate variance AvgVariance are: Where m represents the number of points in each straight line segment; n represents the number of straight line segments in each target; C m Represents the heading information of the mth track point in the track; Calculate the heading change value for the target's low-speed track points, with clockwise being positive and counterclockwise being negative, and take the average value of the overall heading change; The variance of the azimuth change rate of the low-speed point is to directly calculate the azimuth instead of the heading value to make the track shape feature obvious. The track data used is T = {t1, t2, ..., t n }, the heading value is calculated using the azimuth in degrees; The speed variance of the low-speed point is T" = {L1, L2, ..., L n }Calculate the variance of the speed; The maximum distance of the continuous points in the low-speed section is T" = {L1, L2, ..., L n }Calculate the distance between the two farthest points in the track, calculate each low-speed straight track segment separately, and take the maximum value; The ratio of the maximum distance of the point track to the cumulative navigation length is the ratio of the distance between the two farthest points in the target track to the target cumulative navigation length; The ratio of the maximum distance of the point track to the number of track points is the ratio of the distance between the two farthest points in the target track to the number of track points, and the average distance between every two points of the target is calculated.

10. The ship gillnet fishing behavior identification system based on track spatiotemporal characteristics according to claim 6, characterized in that: In the module M5, the geographical information and related configuration data of the fishing area are determined in combination with the environmental factors in the actual project, and the recognition logic of the gillnet fishing target is adjusted; The training method of the decision tree model belongs to a white box model, and the training logic of the decision tree model for gillnet fishing target analysis is incorporated into the code; The module M6 stores the target track data and feature data identified by the gillnet fishing recognition algorithm in a database, and issues a warning to the identified gillnet target in the prohibited fishing area.

Images