Vessel trajectory classification via transfer learning with Deep Convolutional Neural Networks.

Area of Science:

• Maritime safety and navigation
• Artificial intelligence in transportation
• Data analysis and machine learning

Background:

• Automatic Identification System (AIS) data is vital for maritime safety and efficient ship navigation.
• Deep learning, particularly Convolutional Neural Networks (CNNs), offers advanced methods for classifying vessel trajectories.
• Current CNN approaches often struggle to capture nuanced features distinguishing sailing from loitering patterns.

Purpose of the Study:

• To develop a novel deep convolutional neural network (DCNN) framework for simultaneous extraction and classification of both sailing and loitering vessel trajectories.
• To improve the accuracy and efficiency of vessel trajectory classification by capturing subtle differences in movement patterns.
• To leverage transfer learning to reduce data dependency and mitigate overfitting issues.

Main Methods:

• Introduction of the Dense121-VMC framework, a DCNN utilizing transfer learning.
• Application of the framework to extract significant features from input images representing vessel trajectories.
• Simultaneous classification of both sailing and loitering movement patterns.

Main Results:

• The Dense121-VMC framework demonstrates efficiency in extracting key features from vessel trajectory data.
• The approach successfully identifies subtle differences between sailing and loitering trajectories.
• Transfer learning effectively reduced the need for extensive datasets and helped prevent overfitting.

Conclusions:

• The proposed Dense121-VMC framework offers a novel and effective solution for vessel trajectory classification.
• This method significantly contributes to enhancing maritime safety and navigation efficiency.
• The framework's ability to handle complex trajectory patterns showcases its potential for real-world applications.