A Semi-Supervised Methodology for Fishing Activity Detection Using the Geometry behind the Trajectory of Multiple...

Area of Science:

• Maritime surveillance
• Data science
• Machine learning

Background:

• Automatic Identification System (AIS) data is crucial for global vessel tracking.
• Identifying specific vessel activities like fishing from AIS data remains challenging.

Purpose of the Study:

• To develop a semi-supervised, geometry-driven method for detecting fishing activities using AIS data.
• To improve the accuracy and efficiency of identifying fishing patterns from vessel trajectories.

Main Methods:

• Extracting route geometry features from AIS messages.
• Utilizing unsupervised cluster analysis for trajectory labeling.
• Applying recurrent neural networks (RNNs) for time-series classification of fishing activities.

Main Results:

• Achieved an F-score of approximately 87% for fishing activity detection on unseen fishing vessels.
• Demonstrated the effectiveness of the geometric-driven approach in analyzing vessel mobility patterns.
• Conducted a benchmark study comparing various RNN architectures.

Conclusions:

• The proposed method offers a robust pipeline for data preparation, labeling, modeling, and validation.
• This work presents a novel solution for mobility pattern detection by analyzing trajectory geometry.
• The findings contribute to enhanced maritime surveillance and fisheries management through advanced data analysis.