Ship type identification method and device based on motion characteristics of track point set

Abstract

The application provides a ship type identification method and device based on motion characteristics of a track point set, which comprises the following steps: collecting track point data of a ship by a ship positioning device, and preprocessing the track point data; constructing a track point set according to the preprocessed track point data, and extracting motion characteristics of the ship from the track point set; and identifying the ship type by using a deep learning algorithm according to the motion characteristics of the ship, and obtaining the ship type. The application has high accuracy and stability, and has good expansibility.

Description

Technical Field

[0001] This invention relates to the field of maritime information technology, and in particular to a method and apparatus for identifying ship types based on the motion characteristics of track point sets. Background Technology

[0002] With the rapid development of the maritime industry and the continuous growth in the number of ships, the need for ship supervision and identification is becoming increasingly urgent. Traditional ship identification methods mainly rely on Automatic Identification Systems (AIS). While these systems can provide basic information about ships, their identification effectiveness is often limited in certain situations (such as when AIS signals are lost or interfered with), resulting in insufficient accuracy and stability in ship type identification. Furthermore, existing methods largely focus on the analysis of static information, making accurate identification difficult when ships have similar appearances. Summary of the Invention

[0003] This invention provides a method and apparatus for ship type identification based on the motion features of trackpoint sets, which addresses the shortcomings of existing ship type identification technologies, such as insufficient accuracy, poor stability, and strong dependence on AIS signals. By analyzing ship trackpoint data, extracting its motion features, and combining deep learning algorithms, it achieves high-precision ship type identification with high accuracy and stability. It also has good scalability and can be extended to other ship-related tasks, including but not limited to ship behavior analysis and ship trajectory prediction.

[0004] This invention provides a method for ship type identification based on the motion characteristics of track point sets, comprising:

[0005] Ship track point data is collected using ship positioning equipment, and the track point data is preprocessed.

[0006] A set of track points is constructed based on the preprocessed track point data, and the motion characteristics of the ship are extracted from the set of track points.

[0007] Based on the motion characteristics of the vessel, a deep learning algorithm is used to identify the vessel type and obtain the vessel type.

[0008] According to the present invention, a ship type identification method based on the motion characteristics of a set of waypoints is provided, wherein the waypoint data includes one or more of the following: ship MMSI, latitude and longitude, timestamp, speed above ground (SOG), and heading above ground (COG).

[0009] According to the present invention, a ship type identification method based on the motion characteristics of track point sets is provided, wherein the preprocessing includes one or more of the following: data cleaning, outlier processing, duplicate value removal, time series processing, data imputation, and coordinate transformation.

[0010] According to the present invention, a ship type identification method based on the motion characteristics of a trackpoint set is provided, which constructs a trackpoint set based on preprocessed trackpoint data, including:

[0011] Regularize the velocity in the waypoint data;

[0012] The waypoint data are grouped according to the ship's MMSI;

[0013] Each set of waypoint data is converted into image data.

[0014] According to the present invention, a ship type identification method based on the motion characteristics of a track point set is provided, wherein the motion characteristics of the ship include one or more statistical characteristics among ground speed (SOG), ground heading (COG), acceleration, angular velocity, and the entire motion trajectory.

[0015] The statistical characteristics of the ground velocity SOG include one or more of the mean, standard deviation, maximum value and minimum value of the ground velocity SOG;

[0016] The statistical characteristics of the ground heading COG include one or more of the mean of the sine and cosine values ​​of the ground heading COG and the variance of the sine and cosine values ​​of the ground heading COG.

[0017] The statistical characteristics of the acceleration include the mean and / or maximum value of the acceleration;

[0018] The statistical characteristics of the angular velocity include the mean and / or maximum value of the angular velocity;

[0019] The statistical characteristics of the entire trajectory include the average displacement distance and / or the total distance traveled.

[0020] According to the present invention, a ship type identification method based on the motion characteristics of a trackpoint set is provided. Based on the motion characteristics of the ship, a deep learning algorithm is used to identify the ship type, resulting in the ship type, including:

[0021] The image data is divided into multiple grid units;

[0022] Extract the feature map of the ship from each grid cell;

[0023] Based on the motion features and the feature map corresponding to each grid cell, the deep learning algorithm is used to predict multiple bounding boxes of different scales for each grid cell, and outputs the class probability, coordinate information and targetability score.

[0024] According to the ship type identification method based on the motion features of track point sets provided by the present invention, the loss function for training the deep learning algorithm is as follows:

[0025]

[0026] Where DIOU is the loss function, and IOU is the sum of the predicted bounding box b and the ground truth bounding box b. gt The intersection and union ratio, This represents the distance between the center points of two bounding boxes. This represents the diagonal length of the training sample image data.

[0027] According to the present invention, a ship type identification method based on the motion characteristics of a trackpoint set is provided. Based on the motion characteristics of the ship, a deep learning algorithm is used to identify the ship type, resulting in the ship type, including:

[0028] Evaluate the performance metrics of various deep learning algorithms on a validation set;

[0029] The scores for each deep learning algorithm are obtained by weighted summing of multiple performance metrics.

[0030] The score of each deep learning algorithm is multiplied by a preset smoothing factor to obtain the weight of each deep learning algorithm.

[0031] Based on the motion characteristics of the ship, each deep learning algorithm is used to identify the ship type, and the probability of each ship type is obtained;

[0032] The probabilities of each ship type obtained by each deep learning algorithm are weighted and summed, and the ship type with the highest probability is taken as the identification result.

[0033] This invention also provides a method for ship type identification based on the motion characteristics of track point sets, including:

[0034] The data acquisition module is used to collect the ship's track point data through the ship positioning equipment and to preprocess the track point data;

[0035] The feature extraction module is used to construct a set of track points based on the preprocessed track point data, and extract the motion features of the ship from the set of track points;

[0036] The type identification module is used to identify the ship type based on the ship's motion characteristics using a deep learning algorithm, thereby obtaining the ship type.

[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ship type identification method based on the motion characteristics of track point sets as described above.

[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the ship type identification method based on the motion characteristics of a set of waypoints as described above.

[0039] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the ship type identification method based on the motion characteristics of a set of waypoints as described above.

[0040] The ship type identification method and apparatus based on the motion characteristics of track point sets provided by this invention have the following main advantages:

[0041] Improved accuracy and stability of identification: By analyzing the motion characteristics of the ship's track point set, this invention can achieve high-precision identification of ship type, effectively avoiding the identification failure problem caused by AIS signal interference or loss in traditional methods.

[0042] Enhanced system robustness: An improved DIOU loss function is adopted to optimize model performance, which significantly mitigates the impact of bounding box localization errors on target recognition and further improves the stability and reliability of the system.

[0043] Highly scalable and widely applicable: The technical framework of this invention is flexible and universal, and can be easily extended to other ship-related tasks, such as ship behavior analysis and trajectory prediction, providing strong technical support for intelligent shipping and ship supervision. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the ship type identification method based on the motion characteristics of track point sets provided by the present invention.

[0046] Figure 2 This is a schematic diagram of the structure of the ship type identification device based on the motion characteristics of track point sets provided by the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0048] The following is combined Figure 1 A method for ship type identification based on the motion characteristics of track point sets according to the present invention includes:

[0049] Step 101: Collect the ship's track point data through the ship positioning equipment, and preprocess the track point data;

[0050] Step 102: Construct a set of track points based on the preprocessed track point data, and extract the motion characteristics of the ship from the set of track points;

[0051] Step 103: Based on the motion characteristics of the ship, a deep learning algorithm (such as the improved YOLOv5) is used to identify the ship type to obtain the ship type.

[0052] The ship's track point data is collected in real time through ship positioning equipment (such as AIS or GPS), including information such as latitude and longitude, timestamp, speed, and heading.

[0053] The collected data undergoes preprocessing, including data cleaning, outlier handling, duplicate removal, time series data processing, and data imputation, to ensure data accuracy and consistency.

[0054] Based on the preprocessed trackpoint data, a trackpoint set for the ship is constructed. The trackpoint set includes all trackpoints of the ship within a certain period of time, and can comprehensively reflect the ship's motion trajectory.

[0055] The motion characteristics of ships are extracted from track points, including but not limited to speed, heading, acceleration, and angular velocity. These characteristics reflect the motion state and patterns of ships and are an important basis for identifying ship types.

[0056] Based on the extracted motion features, deep learning algorithms (such as the improved YOLOv5) are used to identify ship types. A classifier model is constructed, taking the extracted motion features as input and outputting the ship type.

[0057] Existing methods primarily focus on the analysis of static information, lacking in-depth exploration of the dynamic motion characteristics of ships. Therefore, there is an urgent need for a ship type identification technology based on the motion characteristics of track point sets to improve the accuracy and stability of identification and overcome the limitations of traditional methods. This technology can not only effectively address AIS signal issues but also provide new solutions for tasks such as ship behavior analysis and trajectory prediction.

[0058] This embodiment significantly improves the accuracy and stability of ship type identification by analyzing the motion characteristics of ship track point sets, while avoiding identification failures caused by AIS signal interference or loss in traditional methods. Furthermore, the technical framework of this embodiment has good scalability and can be widely applied to related tasks such as ship behavior analysis and trajectory prediction.

[0059] Based on the above embodiments, the waypoint data in this embodiment includes one or more of the following: ship MMSI, latitude and longitude, timestamp, speed on ground (SOG), and heading on ground (COG).

[0060] The original AIS message can be obtained from the data source and parsed to extract fields such as MMSI (ship identification code), latitude and longitude, timestamp, SOG, and COG.

[0061] Preprocessing includes: deleting outliers with an SOG greater than 30 knots (considered abnormally high speed) or whose latitude and longitude are clearly over land. Secondly, trackpoints with the same MMSI are sorted by timestamp and duplicates are removed. Finally, for point pairs with time intervals exceeding 10 minutes, linear interpolation is performed to standardize the trackpoint time intervals to 1 minute.

[0062] Based on the above embodiments, the preprocessing in this embodiment includes one or more of the following: data cleaning, outlier handling, duplicate value removal, time series sorting, data imputation, and coordinate transformation.

[0063] Data cleaning involves cleaning the collected data to remove null values, outliers, duplicate values, and out-of-series data to ensure data integrity and consistency.

[0064] Data imputation: Interpolation is used to supplement the missing data after cleaning, so as to avoid analysis errors caused by missing data.

[0065] Coordinate transformation: Convert the collected geographic coordinates into a unified projected coordinate system to facilitate subsequent processing.

[0066] Based on the above embodiments, this embodiment constructs a waypoint set according to the preprocessed waypoint data, including:

[0067] Regularize the velocity in the waypoint data;

[0068] The waypoint data are grouped according to the ship's MMSI;

[0069] Each set of waypoint data is converted into image data.

[0070] The preprocessed track point data is converted into image data, so that each ship's track point data corresponds to one image.

[0071] During the conversion process, the velocity of each trajectory point is regularized using the following formula:

[0072]

[0073] Where a is the regularized velocity and v is the original trajectory velocity. The maximum speed in the set of waypoints.

[0074] This embodiment proposes a feature extraction method based on trajectory data regularization and image processing, which significantly improves the expressive power of motion features.

[0075] Based on the above embodiments, the motion characteristics of the ship in this embodiment include one or more statistical characteristics among ground speed SOG, ground heading COG, acceleration, angular velocity, and the entire motion trajectory;

[0076] The statistical characteristics of the ground velocity SOG include one or more of the mean, standard deviation, maximum value and minimum value of the ground velocity SOG;

[0077] The statistical characteristics of the ground heading COG include one or more of the mean of the sine and cosine values ​​of the ground heading COG and the variance of the sine and cosine values ​​of the ground heading COG.

[0078] The statistical characteristics of the acceleration include the mean and / or maximum value of the acceleration;

[0079] The statistical characteristics of the angular velocity include the mean and / or maximum value of the angular velocity;

[0080] The statistical characteristics of the entire trajectory include the average displacement distance and / or the total distance traveled.

[0081] The COG (Course of Gaussian Heading) first calculates the sine and cosine values ​​to handle the jumps between 0° and 360°, and then calculates its mean and variance.

[0082] Acceleration is obtained by dividing the difference in SOG (Situation of Gaussian Governing) between adjacent points by the time difference. Angular velocity is obtained by dividing the difference in COG (Current Coordinate of Gaussian Governing) between adjacent points by the time difference, and by handling circular jumps.

[0083] Based on the above embodiments, this embodiment uses a deep learning algorithm to identify the ship type according to the ship's motion characteristics, and obtains the ship type, including:

[0084] The image data is divided into multiple grid units;

[0085] Extract the feature map of the ship from each grid cell;

[0086] Based on the motion features and the feature map corresponding to each grid cell, the deep learning algorithm is used to predict multiple bounding boxes of different scales for each grid cell, and outputs the class probability, coordinate information and targetability score.

[0087] Ship type identification can employ a YOLOv5 network architecture, comprising a backbone network, a neck network, and a head network. The backbone network extracts features from the input image; the neck network enhances feature representation; and the head network outputs the classification result.

[0088] The model divides the input image data into multiple grid cells, extracts features from the input image data through a feature extraction network to obtain a feature map of a certain size, and predicts multiple bounding boxes of different scales for each grid cell based on the feature map of each grid cell and the motion features extracted from the track point set, and outputs the class probability, coordinate information and target score of each bounding box.

[0089] Among them, the category probability represents the probability that the target within the bounding box belongs to a certain type of ship; the coordinate information includes the coordinates of the center point of the bounding box, the width, and the height; the targetability score is used to measure the confidence level of whether the bounding box contains a target object.

[0090] Through a multi-scale prediction mechanism, the model can adapt to ship trajectory images of different sizes, thereby improving the robustness of recognition.

[0091] Based on the above embodiments, the loss function for training the deep learning algorithm in this embodiment is:

[0092]

[0093] Where DIOU is the loss function, and IOU is the sum of the predicted bounding box b and the ground truth bounding box b. gt The intersection and union ratio, This represents the distance between the center points of two bounding boxes. This represents the diagonal length of the training sample image data.

[0094] In addition, the loss for each bounding box can be calculated based on the class probability, coordinate information, and target score. The final loss is determined based on the loss for each bounding box's class probability, coordinate information, and target score, as well as the DIOU loss.

[0095] The model is trained using a known set of ship track points as training data. During model training, the model is trained using a known set of ship track points, and the improved DIOU (Distance-IoU) loss function in the above formula is used to optimize model performance. DIOU introduces the constraint of the distance between the center points of the bounding box on the basis of the traditional IoU (Intersection over Union), which effectively alleviates the impact of positioning error on target recognition performance.

[0096] This embodiment introduces an optimized DIOU (Distance-IoU) loss function during model training, which effectively alleviates the impact of bounding box localization error of the traditional IoU loss function on target recognition, thereby improving the model's recognition accuracy and robustness.

[0097] Based on the above embodiments, this embodiment uses a deep learning algorithm to identify the ship type according to the ship's motion characteristics, and obtains the ship type, including:

[0098] Evaluate the performance metrics of various deep learning algorithms on a validation set;

[0099] The scores for each deep learning algorithm are obtained by weighted summing of multiple performance metrics.

[0100] The score of each deep learning algorithm is multiplied by a preset smoothing factor to obtain the weight of each deep learning algorithm.

[0101] Based on the motion characteristics of the ship, each deep learning algorithm is used to identify the ship type, and the probability of each ship type is obtained;

[0102] The probabilities of each ship type obtained by each deep learning algorithm are weighted and summed, and the ship type with the highest probability is taken as the identification result.

[0103] Accuracy, recall, and mAP (mean average precision) can be used as evaluation metrics to comprehensively measure the model's recognition performance.

[0104] The preprocessed data can be grouped by MMSI, with each MMSI corresponding to a complete trajectory. To increase the sample size, each long trajectory is divided into trajectory segments of fixed length 30 points (approximately 30 minutes of data), with adjacent segments having 50% overlap. For each 30-point trajectory segment, motion features are calculated, and each trajectory segment is ultimately represented as an approximately 20-dimensional feature vector. The dataset is then divided into training, validation, and test sets in a 7:2:1 ratio.

[0105] The model performance was evaluated on the validation set, and the weighted F1 score of each model was calculated. A smoothing factor was set. The weights of each performance index are calculated.

[0106] For a single sample in the test set, multiple models output probability vectors respectively. After fusion according to the weights mentioned above, the type corresponding to the highest probability is taken as the recognition result.

[0107] According to the test results, the fusion model of this embodiment achieved an overall accuracy of 93.2% and an F1 score of 92.5% on the test set.

[0108] The following describes the ship type identification device based on the motion characteristics of track point sets provided by the present invention. The ship type identification device based on the motion characteristics of track point sets described below can be referred to in correspondence with the ship type identification method based on the motion characteristics of track point sets described above.

[0109] like Figure 2 As shown, the device includes a data acquisition module 201, a feature extraction module 202, and a type recognition module 203, wherein:

[0110] The data acquisition module 201 is used to collect the ship's track point data through the ship positioning equipment and to preprocess the track point data;

[0111] The feature extraction module 202 is used to construct a set of track points based on the preprocessed track point data, and extract the motion features of the ship from the set of track points;

[0112] The type identification module 203 is used to identify the ship type by using a deep learning algorithm based on the ship's motion characteristics, and thus obtain the ship type.

[0113] This embodiment significantly improves the accuracy and stability of ship type identification by analyzing the motion characteristics of ship track point sets, while avoiding identification failures caused by AIS signal interference or loss in traditional methods. Furthermore, the technical framework of this embodiment has good scalability and can be widely applied to related tasks such as ship behavior analysis and trajectory prediction.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship type recognition method based on motion features of a set of track points, characterized in that, The method comprises: Collecting ship track point data of a ship by a ship positioning device, and preprocessing the track point data; Constructing a track point set according to the preprocessed track point data, and extracting motion features of the ship from the track point set; Identifying the ship type by using a deep learning algorithm according to the motion features of the ship, and obtaining the ship type; Constructing a track point set according to the preprocessed track point data, comprising: Regularizing the speed in the track point data; Grouping the track point data according to the ship MMSI; Converting each group of track point data into image data; Identifying the ship type by using a deep learning algorithm according to the motion features of the ship, comprising: Dividing the image data into multiple grid units; Extracting feature maps of the ship from each grid unit; According to the motion features and the feature maps corresponding to each grid unit, the deep learning algorithm is used to predict multiple boundary boxes of different scales for each grid unit, and the class probability, coordinate information and target score are outputted; The loss function for training the deep learning algorithm is: ； Wherein, DIOU is a loss function, IOU is the intersection over union of the predicted bounding box b and the real bounding box b gt , represents the distance between the centers of two bounding boxes, represents the diagonal length of the training sample image data.

2. The ship type identification method based on motion features of a set of track points according to claim 1, characterized in that, The track point data includes one or more of ship MMSI, latitude and longitude, timestamp, ground speed SOG and ground course COG.

3. The ship type identification method based on motion features of a set of track points according to claim 1, characterized in that, The preprocessing includes one or more of data cleaning, outlier processing, repeated value removal, time series arrangement, data filling operation and coordinate conversion.

4. The ship type recognition method based on motion features of a set of track points according to claim 1, characterized in that, The motion features of the ship include one or more of the statistical features of the ground speed SOG, the ground course COG, the acceleration, the angular velocity and the entire motion trajectory; The statistical features of the ground speed SOG include one or more of the mean, the standard deviation, the maximum value and the minimum value of the ground speed SOG; The statistical features of the ground course COG include one or more of the mean of the sine and cosine values of the ground course COG, and the variance of the sine and cosine values of the ground course COG; The statistical features of the acceleration include the mean and / or the maximum value of the acceleration; The statistical features of the angular velocity include the mean and / or the maximum value of the angular velocity; The statistical features of the entire motion trajectory include the average displacement distance and / or the total distance.

5. The ship type recognition method based on motion features of a set of track points according to claim 1, characterized in that, Identifying the ship type by using a deep learning algorithm according to the motion features of the ship, comprising: Evaluating the performance indicators of multiple deep learning algorithms on the validation set; Weighted adding the multiple performance indicators of each deep learning algorithm as the score of each deep learning algorithm; Multiplying the score of each deep learning algorithm by a preset smoothing factor to obtain the weight of each deep learning algorithm; Identifying the ship type by using each deep learning algorithm according to the motion features of the ship to obtain the probability of each ship type; Weighting and adding the probability of each ship type obtained by each deep learning algorithm using the weight of each deep learning algorithm, and taking the ship type with the maximum probability as the identification result.

6. A ship type recognition device based on motion features of a set of track points, characterized in that, The method is applied to the ship type identification method based on the motion features of the track point set according to any one of claims 1 to 5, comprising: The data acquisition module is configured to collect track point data of the ship by using a ship positioning device and to pre-process the track point data. The feature extraction module is configured to construct a track point set according to the pre-processed track point data and to extract motion features of the ship from the track point set. The type identification module is configured to identify the type of the ship by using a deep learning algorithm according to the motion features of the ship, so as to obtain the type of the ship.

7. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the ship type identification method based on the motion features of the track point set according to any one of claims 1 to 5 when executing the program.

Images