Sea Surface Weak Target Detection Based on Non-Linear Multi-Features Fusion

Aiming at the insufficient performance of traditional feature-based detectors and the difficulty of designing high-dimensional detectors, a non-linear multi-features fusion detector is proposed. An accelerated t-SNE algorithm based on the Dual-Tree structure is proposed to effectively reduce the dimensionality of multi-features while preserving the non-linear structure among the original high dimensional features of sea clutter. The experimental results based on real datasets from IPIX and CSIR radars show that the proposed detector further enhances the performance by combining fused non-linear features distribution characteristics. The proposed detector improves the detection probability by more than 8% on average, and exhibits a satisfied performance even under a shorter observation time.