LoRA增强对比学习与SAS视觉Transformer

AI 摘要

该研究提出一种面向合成孔径声呐（SAS）自动目标识别（ATR）的三阶段参数高效适配框架，将基于自然图像预训练的 DINOv3 ViT-L 模型迁移到水下声学场景。第一阶段用 LoRA 冻结主干并适配声学域，第二阶段用难负样本挖掘的 Refiner 强化决策边界，第三阶段用监督对比学习（SupCon）拉近目标、远离杂波。在海上 SAS 数据上按任务级地理划分评估发现，只有 LoRA 适配带来显著提升（AUPRC 提升超过一倍），而后续两个阶段与各自匹配对照组相比均无显著差异，作者据此认为高效适配一步已足够，叠加的课程学习并无收益。

正文节选

LoRA Enhanced Contrastive Learning with SAS Vision Transformers Thanks: This work was supported by the Office of Naval Research (N0001426GI00772) and their Internal Applied Research program. Abstract Automatic target recognition (ATR) with synthetic aperture sonar (SAS) enables advanced naval capabilities, but deep learning approaches remain constrained by human-in-the-loop assessment and by the extreme scarcity of imaged targets against background clutter. We present a three-stage parameter-eff

发布时间：2026-09-22 12:00

抓取时间：2026-09-21 12:12

来源机构：arXiv