Language-Prompted Visual Query Learning for Ship–Sea–Land Disambiguation in Maritime Panoptic Segmentation From SAR Imagery

作者
Rong Gui Deng,Tianwen Zhang,Gui Gao,Xiao Ke,Xiaowo Xu,Shunjun Wei,Xiaoling Zhang
出处
期刊:IEEE Transactions on Circuits and Systems for Video Technology [Institute of Electrical and Electronics Engineers]
卷期号:36 (11): 16334-16347 被引量:16
标识
DOI:10.1109/tcsvt.2026.3712585
摘要

Maritime synthetic aperture radar (SAR) image interpretation requires a unified understanding of ship instances and surrounding sea-land regions under complex scattering and interference conditions. Panoptic segmentation provides a natural framework for this task by jointly modeling instance-level ship perception and region-level scene parsing. However, maritime SAR panoptic segmentation remains insufficiently explored, and existing methods mainly rely on visual supervision, which provides limited semantic guidance for resolving severe ship-sea-land ambiguity in cluttered nearshore scenes. To address this problem, we propose a language-prompted visual query learning framework, termed LPVQ, for ship-sea-land disambiguation in maritime SAR panoptic segmentation. Specifically, we develop a language-prompted query formulation (LPQF) to transform image-aligned structured language attributes into ambiguity-aware language queries and transfer their semantics to the visual query space during training, thereby improving the discriminability of ship, sea, and land queries. We further design a progressive spurious-response suppression (PSS) to progressively attenuate ambiguity-related responses and reinforce prediction-consistent semantics across successive decoding stages. By combining language-guided query learning with response-level semantic purification, LPVQ improves unified parsing of ship instances and surrounding sea-land regions under complex scattering and interference conditions, while still preserving a purely visual inference pipeline at test time. Extensive experiments on SAR-SSP, SPSD, and HRSID-PAN demonstrate that LPVQ consistently outperforms state-of-the-art methods, with PQ gains of 2.37%, 1.51%, and 0.86%, respectively. Large-scene evaluations further show that LPVQ learns transferable visual queries and generalizes well across diverse clutter, scattering, and scene conditions.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
strike发布了新的文献求助10
刚刚
mm关注了科研通微信公众号
刚刚
刚刚
andy完成签到 ,获得积分10
1秒前
简默完成签到,获得积分10
1秒前
秋风的应助被李小鱼采纳,获得10
2秒前
飞飞翔的小马完成签到,获得积分20
2秒前
才高八斗发布了新的文献求助10
2秒前
2秒前
2秒前
2秒前
星星发布了新的文献求助10
2秒前
weijun完成签到,获得积分10
2秒前
喵了个咪完成签到,获得积分10
3秒前
芝麻关注了科研通微信公众号
3秒前
一一完成签到,获得积分10
3秒前
4秒前
张子文完成签到,获得积分10
4秒前
hehe完成签到,获得积分10
4秒前
LXZ发布了新的文献求助10
4秒前
郝岩完成签到,获得积分10
4秒前
4秒前
5秒前
lvy发布了新的文献求助10
5秒前
6秒前
6秒前
6秒前
6秒前
羽宇完成签到,获得积分10
6秒前
xmn发布了新的文献求助10
6秒前
张子文发布了新的文献求助10
6秒前
xiuxiu酱发布了新的文献求助10
7秒前
8秒前
9秒前
lgh完成签到,获得积分10
9秒前
菖蒲完成签到,获得积分10
9秒前
李爱国的应助被洛枝采纳,获得10
9秒前
吕不好发布了新的文献求助10
10秒前
吕不好发布了新的文献求助10
10秒前
吕不好发布了新的文献求助10
10秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
奈特基础组织学:配套组织病理学(第4版)(Netter's Essential Histology With Correlated Histopathology) 800
John McDowells Bildungstheorie 800
八旗制度の研究 600
plantTFDB 508
Resource Recovery Handbook 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8175853
求助须知:如何正确求助?哪些是违规求助? 9567457
关于积分的说明 21792632
捐赠科研通 7770447
什么是DOI,文献DOI怎么找? 3431918
关于科研通互助平台的介绍 2556276
邀请新用户注册赠送积分活动 2477418