Fishing operation type recognition based on multi-branch convolutional neural network using trajectory data.

Area of Science:

• Maritime surveillance
• Spatial computing
• Fisheries science

Background:

• Sustainable fishery management requires accurate identification of fishing vessel operations.
• Current methods struggle with spatiotemporal context and multimodal data fusion in vessel trajectories.

Purpose of the Study:

• To develop a novel framework for enhanced fishing vessel operation recognition using spatiotemporal trajectory data.
• To improve the fusion of multimodal data and extraction of contextual information.

Main Methods:

• Integration of Geohash-based geocoding with natural language processing-inspired embedding techniques for spatiotemporal feature extraction.
• Development of a multi-branch 1D convolutional neural network (MB-1dCNN) for operational-type recognition.
• Comparative analysis of Geohash encoding lengths and network architectures (single-branch vs. multi-branch, fully-connected vs. 1D-CNN).

Main Results:

• Optimal Geohash encoding length determined to be 5.
• Multi-branch architectures significantly outperformed single-branch designs.
• The MB-1dCNN achieved superior accuracy and F1-score compared to multi-branch fully-connected networks (MB-FCNN).

Conclusions:

• 1D-CNN processing is more effective for sequential feature extraction than fully-connected networks.
• Multi-branch architectures enhance information fusion capabilities in trajectory analysis.
• The MB-1dCNN model sets a new state-of-the-art for trajectory-based fishing operation recognition.