Early Intent Recognition for Maritime Domains
Abstract
Two papers accepted and presented by coauthors. The work covers feature-aware deep learning models that achieve ~97% accuracy on seven maritime behaviors using only initial trajectory segments, and rolling-threat prediction with 81–88% multiclass accuracy using short 20-second observation windows. The presentations highlighted how temporal ML/DL models enable early intent recognition despite sparse and noisy observations—critical for naval situational awareness and decision support.
Date
Aug 17, 2025 9:00 AM — Aug 21, 2025 5:00 PM
Event
Location
Los Angeles, California, USA
506 South Grand Avenue, Los Angeles, California 90071
Two papers accepted and presented by coauthors at IEEE CASE 2025 (Los Angeles) on maritime intent recognition:
Early Classification of Intentions for Maritime Domains Using Deep Learning Models — Feature-aware LSTM, Bi-LSTM, and Transformer classifiers for early vessel intent prediction.
Proactive Maritime Threat Prediction: Vessel Intent Classification with LSTMs and Transformers Using a Sliding Window Approach — Rolling-threat prediction with sliding-window formulation for real-time situational awareness.
