Similarity metric based feature fusion for active sonar target classification

We propose a novel multi-feature fusion strategy that selectively merges hand-crafted acoustic features with low inter-feature similarity to enhance active sonar classification in data-scarce scenarios. While deep learning models typically learn latent representations automatically, the limited availability of active sonar data necessitates reliance on hand-crafted features, which inevitably cause information loss. Our method addresses this limitation by systematically measuring feature similarity through an analysis of common neighbors between two feature sets and fusing only those sets determined to be sufficiently dissimilar. Experimental results demonstrate that the selective fusion of low-similarity feature sets yields better classification performance compared to fusing highly similar features, thereby underscoring the effectiveness of our approach in maximizing complementary information from limited sonar datasets. [This work was supported by Korea Research Institute for defense Technology planning and advancement (KRIT)—Grant funded by the Korea government (DAPA (Defense Acquisition Program Administration)) (No. KRIT-CT-23-026, Integrated Underwater Surveillance Research Center for Adapting Future Technologies, 2025)]