System and method for detecting marine activity

Abstract

The present application provides a method for detecting vessels engaged in illegal fishing, comprising the steps of: receiving data from an automatic identification system (AIS) of a vessel; pre-processing the AIS data to produce a set of motion-related features; constructing one or more trajectory representations from the set of motion-related features; extracting one or more features from the one or more trajectory representations using an autoencoder module; and, selecting one or more random samples of data for use by an artificial intelligence engine to detect vessels engaged in fishing activities. The artificial intelligence engine comprises an unsupervised learning module for labelling the one or more trajectory representations as a plurality of labels, and a supervised learning module for generating a machine learning model to determine whether a vessel is engaged in illegal fishing activities.

Description

Technical Field

[0001] This invention relates to an advanced and innovative semi-supervised model that uses data from the Automatic Identification System (AIS data) to detect fishing activities. Background Technology

[0002] Illegal fishing is a major problem threatening global fisheries, marine biodiversity, ecosystem balance, and fish populations. In Hong Kong, illegal fishing occurs almost daily, severely impacting local fishery resources and marine ecosystems. Under the Fisheries Protection Ordinance, fishing using unregistered vessels or fishing outside of designated times within controlled areas of Hong Kong waters is considered illegal fishing. Therefore, monitoring fishing vessels and detecting their activity patterns is crucial for combating illegal fishing and protecting Hong Kong's marine ecosystem. Maritime authorities worldwide are actively seeking ways to improve the timeliness and efficiency of existing manual pattern analysis processes to better plan and allocate resources and deploy actions to combat illegal fishing in Hong Kong. To assist maritime authorities in achieving this goal, it is necessary to develop an automated and intelligent solution suitable for Hong Kong, capable of effectively detecting fishing activities.

[0003] The development of navigation systems aims to monitor and track fishing vessels, thus aiding in the fight against illegal fishing. Currently, most navigation data comes from Automatic Identification Systems (AIS), a promising data source for identifying, monitoring, and recording vessel movements without radar detection. It allows offshore equipment to exchange information in real time, serving as an effective tool for uncovering hidden fishing patterns with the assistance of artificial intelligence (AI) algorithms. Extensive research is currently focused on developing more advanced AI technologies to utilize AIS data for fishing activity detection. Arasteh S et al. developed a fishing activity classification model based on convolutional neural networks (CNNs) to identify offshore fishing patterns. This model was trained on a labeled dataset published by Global Fishing Watch (GFW) and validated using datasets from the United States and Denmark. This model compares actual fishing activity with vessel reports to determine whether vessels are engaging in illegal fishing within a specific timeframe. De Souza et al. proposed three fishing activity detection methods based on AIS data, classifying them according to fishing gear type. Each method is specialized for a specific vessel type but cannot be applied to other situations. Unlike the methods mentioned above, Martha Dais Ferreira et al. proposed a semi-supervised method based on AIS data streams for fishing activity detection in situations where labeled data is scarce. This method combines unsupervised and supervised models to infer labels for each data point and perform a point-based classification task on an expanded labeled dataset. While this method relies on a point-based classification model and traditional parameters such as ground heading (COG), COG change, ground velocity (SOG), and acceleration, it is computationally expensive and time-consuming.

[0004] Despite significant progress in detecting vessel movement patterns, these methods remain unimplemented in Hong Kong for the following reasons: First, fishing vessels can manipulate AIS information, such as the vessel's unique MMSI (Maritime Mobility Identifier), COG (Course of Course), and SOG (Speed ​​of Surface). Finally, Hong Kong's AIS reports lack fields indicating the activity status (e.g., fishing or navigation). These issues make it easier for fishing vessels to conceal their actual activities and engage in illegal fishing in Hong Kong, resulting in a lack of the tagging data needed to perform fishing detection tasks. Summary of the Invention

[0005] According to a first aspect of the present invention, a method for detecting vessels engaged in illegal fishing is provided, comprising the following steps:

[0006] Receive data from the Automatic Identification System (AIS data);

[0007] Preprocess AIS data to generate a set of motion-related features;

[0008] One or more trajectory representations are constructed from this set of motion-related features;

[0009] The autoencoder module is used to extract one or more features from the one or more trajectory representations; and

[0010] Select one or more random sample data points for the artificial intelligence engine to detect vessels engaged in fishing activities;

[0011] The artificial intelligence engine includes:

[0012] An unsupervised learning module for labeling one or more trajectory representations as multiple labels; and

[0013] A supervised learning module is used to generate machine learning models to determine whether a vessel is engaged in fishing activities.

[0014] In an embodiment of the first aspect, the unsupervised learning module uses the K-means clustering algorithm to classify features in one or more trajectory representations.

[0015] In an embodiment of the first aspect, the K-means clustering algorithm is configured to classify the data into two clusters.

[0016] In an embodiment of the first aspect, one cluster is labeled "fishing" and another cluster is labeled "navigation".

[0017] In an embodiment of the first aspect, the supervised learning module utilizes the random forest (RF) algorithm to generate a machine learning model for determining vessels involved in fishing.

[0018] In an embodiment of the first aspect, the supervised learning module includes a verification module for verifying the correctness of the machine learning model.

[0019] In an embodiment of the first aspect, the supervised learning module includes an evaluation module for evaluating a machine learning model based on one or more evaluation metrics.

[0020] In embodiments of the first aspect, the evaluation metrics include one or more of the following: precision, recall, F1 score, and accuracy.

[0021] In an embodiment of the first aspect, the trajectory representation includes one or more AIS data points.

[0022] In the first aspect of the embodiment, the autoencoder module is implemented using a convolutional neural network.

[0023] In an embodiment of the first aspect, preprocessing the AIS data includes the step of cleaning the original AIS dataset.

[0024] In an embodiment of the first aspect, preprocessing the AIS data includes the step of generating a set of motion-related features, such as recalculated ground speed (SOG), acceleration, and recalculated change in ground heading (COG).

[0025] In an embodiment of the first aspect, the trajectory representation is constructed by combining a map of Hong Kong, China, with a set of ship trajectories formed by motion-related features.

[0026] In an embodiment of the first aspect, the Automatic Identification System (AIS) data is collected from remote sensing technology.

[0027] In an embodiment of the first aspect, the AIS data includes one or more of the following: Maritime Mobility Service Identifier (MMSI), navigation status, rate of turn, ground speed, spatial details (e.g., longitude and latitude), ground heading, and UTC seconds.

[0028] According to a second aspect of the present invention, a system for detecting vessels engaged in illegal fishing activities is provided, wherein the system comprises:

[0029] One or more sensors are used to collect Automatic Identification System (AIS) data from the ship's sensors in real time;

[0030] A ground station server includes one or more central processing units (CPUs), graphics processing units (GPUs), or tensor processing units (TPUs) for performing tensor or multidimensional array calculations and operations in artificial intelligence algorithms, and includes read-only memory (ROM) and random access memory (RAM), wherein the ground station server is adapted to:

[0031] The AIS data is preprocessed to generate a set of motion-related features;

[0032] Construct one or more trajectory representations from this set of motion-related features;

[0033] The autoencoder module extracts one or more features from one or more trajectory representations;

[0034] Select one or more random sample data points for the artificial intelligence engine to detect vessels engaged in fishing activities;

[0035] The artificial intelligence engine includes:

[0036] An unsupervised learning module for labeling one or more trajectory representations as multiple labels; and

[0037] A supervised learning module is used to generate machine learning models to determine whether a vessel is engaged in fishing activities.

[0038] In an embodiment of the first aspect, the step of preprocessing the AIS data includes removing duplicate data, repairing missing AIS data, and removing abnormal data. Attached Figure Description

[0039] Embodiments of the present invention will now be described by way of example and with reference to the accompanying drawings, wherein:

[0040] Figure 1 This is a schematic diagram of a system for detecting vessels engaged in illegal fishing activities in Hong Kong, China, according to one embodiment of the present invention.

[0041] Figure 2 for Figure 1 The diagram shows a process implemented in the system.

[0042] Figure 3 for Figure 1 The diagram shows the overall structural framework of the system implemented for detecting vessels engaged in illegal fishing activities.

[0043] Figure 4 for Figure 1 A schematic diagram of the proposed workflow of the artificial intelligence engine implemented in the system shown.

[0044] Figure 5 Showing from Figure 1 The confusion matrix generated from the data collected in the system shown. Detailed Implementation

[0045] This invention provides an advanced and innovative semi-supervised model that utilizes Automatic Identification System (AIS) data to detect fishing activities. The system and method aim to identify whether vessels are engaged in fishing activities at specific times and locations, thereby assisting maritime authorities in verifying the eligibility of vessels to fish in Hong Kong.

[0046] refer to Figure 1 and Figure 2 This embodiment illustrates an example of the invention. The aim of this embodiment is to provide a system and method for detecting vessels engaged in illegal fishing, particularly suitable for detecting vessels engaged in fishing activities in Hong Kong, China.

[0047] In this example instance, the interface and processor are implemented by a computer with a suitable user interface. This computer can be implemented using any computing architecture, including laptops, tablets, standalone personal computers (PCs), smart devices, Internet of Things (IoT) devices, edge computing devices, client / server architectures, "dumb" terminal / host architectures, cloud-based computing architectures, or any other suitable architecture. The computer device can be appropriately programmed to implement this invention.

[0048] First turn Figure 1 A schematic diagram of a computer system or computer server 100 is shown, arranged to be implemented for use in an advanced and innovative semi-supervised model according to the present invention, utilizing Automatic Identification System (AIS) data to detect fishing activities. This example includes server 100, which includes suitable components necessary to receive, store, and execute appropriate computer instructions. These components may include a processing unit 102, including a central processing unit (CPU) for tensor or multidimensional array calculations or manipulations, a mathematics coprocessor (mathematics processor), a graphics processing unit (GPU) or tensor processing unit (TPU) operation, read-only memory (ROM) 104, random access memory (RAM) 106, and input / output devices such as a disk drive 108, input devices 110 such as Ethernet ports, USB ports, etc., a liquid crystal display, a light-emitting display, or any other suitable display, and a communication link 114. Server 100 may contain instructions that can be incorporated into ROM 104, RAM 106, or disk drive 108 and executed by processing unit 102. Examples include servers, personal computers, terminals, wireless or handheld computing devices, Internet of Things (IoT) devices, smart devices, and edge computing devices. At least one of the multiple communication links can be connected to an external computer network via a telephone line or other types of communication link.

[0049] Server 100 may also include storage devices, such as disk drive 108, which may include solid-state drives (SSDs), hard disk drives (HDDs), optical disc drives, tape drives, or remote or cloud-based storage devices. Server 100 may use a single disk drive, multiple disk drives, or remote storage services. Furthermore, server 100 may be equipped with a suitable operating system 116, which may be stored on the disk drive or in the read-only memory (ROM) of server 100.

[0050] Calculators or computing devices can also provide the necessary computational power to operate or connect machine learning networks, such as neural networks, to achieve various functions and outputs. Neural networks can run locally or entirely or partially through servers or cloud services. The machine learning network can be untrained, partially trained, or fully trained, and / or can be retrained, orchestrated, or upgraded over time.

[0051] Figure 2 This is a block diagram of a system 150 for detecting vessels 152 engaged in illegal fishing. Vessel 152 is equipped with a transceiver that transmits Automatic Identification System (AIS) data. However, it is assumed that the AIS data may be incomplete and contain defects. A ground station server 154 is equipped with sensor technology to collect AIS data transmitted from vessel 152 in real time. Preferably, the sensor technology includes a receiver for receiving image data of the area. After receiving the AIS data, the ground station server 154 transmits the AIS data to a preprocessing module 162 for cleaning and standardization. The cleaned data is then used to reconstruct a trajectory representation 164 of the vessel. During the training phase, the trajectory representation 164 is classified into two categories with explicit labels by an unsupervised learning model 166. The classification data is then transmitted to a supervised learning module 168 to generate a machine learning model 170. During prediction or classification, the trajectory representation 164 is fed to the machine learning model 170 to detect whether the vessel is engaged in illegal fishing activities.

[0052] Therefore, in one embodiment of the present invention, a method for detecting vessels engaged in illegal fishing activities is provided. The method includes the following steps: receiving data from an Automatic Identification System (AIS); preprocessing the AIS data to generate a set of motion-related features; constructing one or more trajectory representations from the set of motion-related features; extracting one or more features from the one or more trajectory representations using an autoencoder module; and selecting one or more random sample data for use by an artificial intelligence engine to detect vessels engaged in illegal fishing activities; wherein the artificial intelligence engine includes: an unsupervised learning module for labeling the one or more trajectory representations as multiple tags; and a supervised learning module for generating a decision tree model to determine whether a vessel is engaged in illegal fishing activities.

[0053] To address the challenges and problems in existing technologies, this invention provides an advanced and innovative semi-supervised artificial intelligence system that utilizes AIS data to detect fishing activities. This invention is applicable to identifying whether a vessel is engaged in fishing or catching activities at a specific time and place, thereby assisting maritime authorities in verifying the eligibility of vessels to fish or catch fish in Hong Kong, China. Furthermore, this invention is suitable for providing assistance in resource allocation and patrol planning to more effectively combat illegal fishing or catching. Figure 3The overall structural architecture of the artificial intelligence engine of the system according to an embodiment of the present invention is shown. Figure 3 This provides a clear roadmap for the artificial intelligence engine of this invention.

[0054] refer to Figure 3 In a higher-level embodiment of the invention, system 300 is designed to retrieve data from a variety of data sources 302. Data sources 302 may include data from: maritime management personnel 312, remote sensing technology 314, fishing vessels 316, or Automatic Identification System (AIS) 318.

[0055] In one instance, the maritime administrator 312 can use a ground station server for data collection. This ground station server is equipped with various remote sensing technologies 314 or different types of transceivers, such as radar, Global Navigation Satellite System (GNSS), Alternative Positioning, Navigation and Timing Solution (APNT), Magnetic Navigation (MAGNAV), absolute addressing utilizing geomagnetic anomalies, cellular networks, Wi-Fi, Bluetooth, and VHF transceivers. Some of these sensors are designed as part of an Automatic Identification System (AIS) 318 to collect short-range coastal tracking system data, which is broadcast directly in real-time by fishing vessel 316 or other nearby vessels.

[0056] Data obtained from the Automatic Identification System (AIS) may include the following: Maritime Mobility Identifier (MMSI), navigation status, rate of turn, surface speed (SOG), spatial details (e.g., latitude and longitude), surface heading (COG), and UTC seconds. It is understood that AIS data may have some issues, such as missing data or attributes. This can be caused by a variety of factors, such as the ship owner deliberately disabling the AIS system to evade tracking, or the impact of adverse weather conditions on data transmission.

[0057] Ground station servers are typically configured to collect all raw AIS data and process it into meaningful and actionable information. This processed information can then be transmitted back to the ship to support its informed decision-making.

[0058] This invention provides an artificial intelligence engine 304 suitable for processing data obtained from data source 302. In a preferred embodiment, the artificial intelligence engine 304 of this invention is integrated into a ground station system that includes a computer server 100 for receiving and processing data from various remote sensing technologies 314 or different types of transceivers in a real-time environment, while also being able to train the artificial intelligence engine 304.

[0059] In one embodiment of the present invention, the calculator server 100 includes an artificial intelligence engine 304, which is configured in step 320 to identify fishing behavior patterns. To identify fishing behavior patterns, the artificial intelligence engine 304 is configured to perform the following steps: data preprocessing 324, trajectory reconstruction 326, performing unsupervised learning 328, and performing supervised learning 330. In a preferred embodiment, the unsupervised learning process 328 uses the K-means algorithm. Furthermore, the supervised learning process 330 preferably employs the Random Forest (RF) algorithm.

[0060] Based on the capabilities of the artificial intelligence engine 304, the present invention is configured to provide an application 306 that includes one or more of the following functions: exploring the behavior patterns of fishing vessels 332, assisting in the detection of suspicious illegal fishing activities 334, detecting fishing activities in regulated areas 336, providing monitoring and law enforcement support 338, and resolving and combating cross-border illegal fishing activities 340.

[0061] The importance of these applications 306 308 is reflected in their contributions to: ensuring the standardization of marine operations 342, promoting the protection of marine ecosystems 344, supporting sustainable development 346, achieving effective fisheries management 348, enhancing maritime safety 350, and managing fisheries distribution 352.

[0062] The artificial intelligence engine 304 of this invention is configured to overcome known difficulties in prior art by utilizing semi-supervised learning methods. Its workflow is as follows: Figure 4 As shown, the process consists of two main steps: (1) clustering the trajectories using an unsupervised learning model and labeling them according to their unique features to expand the labeled data dataset; then (2) using a supervised learning model to train a random forest (RF)-based classifier using the obtained labeled data dataset for fishing activity detection.

[0063] Now for reference Figure 4 The figure illustrates the workflow of the artificial intelligence engine 304 in an example of the present invention.

[0064] In the artificial intelligence engine 304, a data preprocessing module 402 is provided, which is configured to perform the data collection module 410, data cleaning module 412, trajectory reconstruction module 414, and data normalization module 416. The data preprocessing module 402 is configured to clean the raw dataset and reconstruct the ship's trajectory. This data preprocessing module 402 is configured to generate a set of motion-related features, such as recalculated ground speed (SOG), acceleration, and recalculated change of heading (COG), which are used to generate the final ship trajectory. In a preferred embodiment, the data preprocessing module 402 is configured to collect data from various data sources 302, such as AIS data collected from the data collection module 410. According to one embodiment, the data collection module 410 includes an AD / DA digital signal processor (DSP) suitable for collecting analog and digital signals and processing these signals into digital data for subsequent processing. The raw data from the data source 302 is passed from the data collection module 410 to the data cleaning module 412.

[0065] The data cleaning module 412 performs a series of data cleaning processes, such as removing duplicate data, filtering out-of-bounds data, and handling missing data and attributes. The cleaned data is then passed to the trajectory reconstruction module 414 to reconstruct the trajectories of each vessel. In one embodiment, the trajectory reconstruction module 414 processes the cleaned data and ultimately reconstructs the complete trajectories of the vessels. Then, the data standardization module 416 standardizes the data to prepare it for training purposes.

[0066] The trajectory representation module 404 processes the input data to generate a ship trajectory representation 420 for use by subsequent unsupervised learning modules 406 and supervised learning modules 408. This trajectory representation module 404 includes a convolutional autoencoder algorithm module 422 for extracting and learning latent features of the ship trajectory representation 420. In a preferred embodiment, the ship trajectory representation 420 may include one or more images showing the movement patterns of a ship within a specific area. Furthermore, the ship trajectory representation 420 can be constructed by combining data cleaned from the data preprocessing module 412 with satellite imagery. In another embodiment, an artificial intelligence engine may be used to identify or generate and repair any missing data points related to the ship trajectory.

[0067] In a preferred embodiment, the autoencoder algorithm module 422 includes a deep learning artificial intelligence system specifically designed for extracting and analyzing the intrinsic features embedded in the ship trajectory representation 420. The autoencoder algorithm module 422 can be implemented as a convolutional neural network (CNN), utilizing its hierarchical structure to capture spatial and temporal patterns in the trajectory data. This approach allows for a deeper understanding of the pattern order and relationships within the trajectory data. The flexibility of this implementation ensures that the system can adapt to different data representations and complexities, thereby improving the accuracy and efficiency of feature extraction in various scenarios.

[0068] In the unsupervised learning module 406, the present invention utilizes the unsupervised characteristics of clustering technology 432 to divide ship trajectories 420 with similar movement behavior characteristics into two categories (fishing and navigation) based on the features extracted from the trajectory representation, and assigns labels to them.

[0069] Features extracted from vessel trajectories 420 are fed into an unsupervised learning module 406 for label generation. In a preferred embodiment, the unsupervised learning module 406 implements a K-means clustering algorithm. This algorithm classifies the data into two distinct clusters (K=2): fishing and navigation. After the clustering process is complete, the unsupervised learning module 406 assigns labels to the entire data dataset in step 434.

[0070] Subsequently, in step 436, a subset of the labeled dataset is selected through random sampling. This sampled data is then fed into the supervised learning module 408 for optimization and classification model validation, thereby enabling more accurate prediction and classification of ship trajectories in future analyses. This integrated approach combines the advantages of unsupervised and supervised learning techniques, enhancing the system's performance and reliability.

[0071] Now for reference Figure 4 The supervised learning module 408 is shown. Randomly selected samples from the labeled dataset generated in the unsupervised step 406 are used to train a classification model for detecting fishing activity. In step 442, these random samples serve as training data for the supervised learning process.

[0072] In a preferred embodiment, an artificial intelligence-based algorithm—such as machine learning or deep learning methods—is used to construct the classification model. In one embodiment of the invention, a random forest (RF) algorithm is employed in step 444 to generate the classification model. This algorithm utilizes ensemble learning techniques to improve the accuracy and robustness of predictions.

[0073] To ensure the reliability of the model, it is validated in step 446 using an independent dataset selected through random sampling. Then, in step 448, the performance of the random forest (RF) model is evaluated using one or more established evaluation metrics, including precision, recall, F1 score, and accuracy. These metrics provide a comprehensive assessment of the model's effectiveness in classifying fishing or harvesting activities, thereby facilitating further improvements and optimizations to the detection system.

[0074] The present invention has been described in detail above to enable those skilled in the art to understand it clearly. Therefore, it is conceivable that certain modifications and adjustments can be made without departing from the core principles of this disclosure. It should be noted that there are other alternative methods and configurations for implementing the processes and apparatus discussed herein.

[0075] Therefore, the described embodiments are intended as illustrative and not limiting. Those skilled in the art will understand that various modifications can be made to the described details without departing from the fundamental principles of the invention. Therefore, the scope of this disclosure should be defined only by the following claims.

Claims

1. A method for detecting vessels engaged in fishing, characterized in that, Includes the following steps: Receive data from the Automatic Identification System (AIS data); The AIS data is preprocessed to generate a set of motion-related features; Construct one or more trajectory representations from this set of motion-related features; Use an autoencoder module to extract one or more features from the one or more trajectory representations; as well as Select one or more random sample data points for the artificial intelligence engine to detect vessels engaged in illegal fishing activities; The artificial intelligence engine includes: An unsupervised learning module is used to label the one or more trajectory representations as multiple labels; and A supervised learning module is used to generate machine learning models to determine whether a vessel is engaged in fishing activities.

2. The method for detecting vessels engaged in fishing according to claim 1, characterized in that, in, The unsupervised learning module uses the K-means clustering algorithm to classify the features in the one or more trajectory representations.

3. The method for detecting vessels engaged in illegal fishing according to claim 2, characterized in that, in, The K-means clustering algorithm is configured to classify the data into two clusters.

4. The method for detecting vessels engaged in fishing according to claim 3, characterized in that, in, One of the clusters is labeled "fishing" and the other cluster is labeled "navigation".

5. The method for detecting vessels engaged in fishing according to claim 4, characterized in that, in, The supervised learning module utilizes the Random Forest (RF) algorithm to generate a machine learning model for identifying vessels involved in illegal fishing.

6. The method for detecting vessels engaged in fishing according to claim 5, characterized in that, in, The supervised learning module includes a verification module for validating the correctness of the decision tree model.

7. The method for detecting vessels engaged in fishing according to claim 6, characterized in that, in, The supervised learning module includes an evaluation module for evaluating machine learning models based on one or more evaluation metrics.

8. The method for detecting vessels engaged in fishing according to claim 7, characterized in that, in, Evaluation metrics include one or more of the following: precision, recall, F1 score, and accuracy.

9. The method for detecting vessels engaged in fishing according to claim 8, characterized in that, in, The trajectory representation includes one or more image data.

10. The method for detecting fishing vessels according to claim 9, characterized in that, in, The autoencoder module is implemented using a convolutional neural network.

11. The method for detecting fishing vessels according to any one of claims 6 to 10, characterized in that, in, Preprocessing the AIS data includes the step of cleaning the raw AIS dataset.

12. The method for detecting vessels engaged in fishing according to claim 11, characterized in that, in, Preprocessing the AIS data includes the step of generating a set of motion-related features, including recalculated ground speed (SOG), acceleration, and recalculated change in ground heading (COG).

13. The method for detecting vessels engaged in fishing according to claim 12, characterized in that, in, The trajectory representation is constructed by combining a map of Hong Kong, China, with ship trajectories formed by a set of motion-related features.

14. The method for detecting vessels engaged in fishing according to claim 13, characterized in that, in, Automatic Identification System (AIS) data is collected from remote sensing technology.

15. The method for detecting vessels engaged in fishing according to claim 14, characterized in that, in, The AIS data includes one or more of the following: Maritime Mobility Service Identifier (MMSI), navigation status, rate of turn, ground speed, spatial details (e.g., longitude and latitude), ground heading, true heading, true bearing of its own position, and UTC seconds.

16. A system for detecting vessels engaged in illegal fishing, characterized in that, include: One or more sensors are used to collect data from the vessel's sensors in real time for the Automatic Identification System (AIS data); A ground station server includes one or more central processing units (CPUs), graphics processing units (GPUs), or tensor processing units (TPUs) for performing tensor or multidimensional array calculations and operations in artificial intelligence algorithms, and includes read-only memory (ROM) and random access memory (RAM), wherein the ground station server is adapted to: The AIS data is preprocessed to generate a set of motion-related features; Construct one or more trajectory representations from this set of motion-related features; The autoencoder module extracts one or more features from one or more trajectory representations; Select one or more random sample data points for the artificial intelligence engine to detect vessels engaged in illegal fishing activities; The artificial intelligence engine includes: An unsupervised learning module for labeling one or more trajectory representations as multiple labels; and A supervised learning module is used to generate machine learning models to determine whether a vessel is engaged in illegal fishing activities.

17. The system for detecting vessels engaged in illegal fishing according to claim 16, characterized in that, in, The steps for preprocessing AIS data include removing duplicate data, repairing missing AIS data, and removing abnormal data.

Images