Attention-Guided Track-Pulse-Sequence Target Association Network.

Area of Science:

• Maritime surveillance
• Artificial intelligence
• Signal processing

Background:

• Multi-satellite detection is vital for maritime target identification and tracking.
• Fragmented track segments due to satellite revisit patterns and target motion necessitate effective track association.
• Current methods struggle with localization errors and dense, overlapping target scenarios.

Purpose of the Study:

• To propose an attention-guided track-pulse-sequence target association network (AG-TPS-TAN) for improved multi-satellite track association.
• To overcome limitations of existing methods relying solely on track information or statistical signal parameters.
• To enhance the accuracy and robustness of maritime target tracking.

Main Methods:

• Developed an asymmetric dual-branch network processing both track and raw electromagnetic pulse sequence data.
• Incorporated a track-point-aware attention mechanism to identify critical motion points.
• Utilized a dual-feature fusion module with a combined loss function for enhanced feature consistency and discriminability.

Main Results:

• The AG-TPS-TAN achieved high association accuracies: 93.91% for 5 targets and 63.83% for 50 targets.
• Outperformed track-only (TSADCNN) and signal-statistics-based methods significantly, especially with increasing target numbers.
• Demonstrated superior performance in complex scenarios with dense targets and overlapping parameters.

Conclusions:

• The proposed AG-TPS-TAN effectively addresses limitations in current multi-satellite track association methods.
• The novel approach shows significant advantages in accuracy and robustness for maritime target identification and tracking.
• This work advances the field of AI-driven maritime surveillance and signal processing.