Lightweight SAR Ship Detection via Contrastive Distillation

2026-05-27

AI Summary

Existing lightweight SAR ship detection models struggle to capture the complex structural relationships in radar echoes, while conventional knowledge distillation methods focus solely on local features or logits, neglecting the geometric correlations among object representations. To address this limitation, this work proposes SURGE, the first Transformer-based knowledge distillation framework for SAR ship detection. SURGE efficiently transfers relational geometric knowledge from a teacher model to a student model via a contrastive InfoNCE loss within a shared projected embedding space. The method introduces an architecture-agnostic, region-level distillation interface compatible with two-stage, one-stage, and Transformer-based detectors. Extensive experiments on SSDD and HRSID demonstrate significant performance gains, achieving up to 6.2 mAP and 8.0 AP75 improvements—surpassing even the teacher model in certain metrics.

Abstract

Deep convolutional and transformer-based detectors achieve strong performance for SAR ship detection but are often computationally prohibitive for real-time or onboard deployment. Lightweight models offer improved efficiency yet struggle to capture the complex structural relationships inherent in SAR backscatter. Most existing SAR knowledge-distillation approaches rely on feature or logit matching, which enforces localized activation similarity while neglecting the geometric relationships among object representations. We propose a Structured Unified Relational knowledGE distillation framework for SAR Ship detection (SURGE) that transfers relational geometry from a powerful teacher detector to a compact student detector using a contrastive InfoNCE objective in a shared projection embedding space. To the best of our knowledge, this work presents the first transformer-based SAR ship detector knowledge distillation framework in SAR domain. The framework is architecture-agnostic in the sense that it provides a common region-level distillation interface for two-stage, one-stage and transformer-based detectors without modifying their deployed architectures. Experiments on the SSDD and HRSID benchmarks demonstrate that the proposed method yields substantial improvements for two-stage detectors, achieving up to 6.2 mAP and 8.0 AP75 gains over baseline student and even surpassing teacher performance

Problem

• SAR ship detection
• lightweight models
• knowledge distillation
• relational geometry
• structural relationships

Innovation

• knowledge distillation
• contrastive learning
• SAR ship detection
• relational geometry
• lightweight model