MambaDSF: Multi-Scale SSM with Dilated Feature Fusion for Sonar Small Target Detection
Researchers have developed MambaDSF, a novel framework for detecting small targets in sonar imagery. This hybrid approach combines a Mamba-based backbone with dilated feature fusion to efficiently capture both local and global acoustic context. The system introduces new loss functions to improve training stability for small targets and has demonstrated superior performance on the UATD benchmark, outperforming existing detectors. AI
IMPACT Introduces a new architecture for underwater target detection, potentially improving autonomous underwater vehicle capabilities.