Litcius/Paper detail

Tri-State Prototype Self-Distillation for SAR Ocean Imagery Panoptic Segmentation

Rong Gui Deng, Tianwen Zhang, Xiaowo Xu, Xiaoling Zhang, Gui Gao

2026IEEE Geoscience and Remote Sensing Letters6 citationsDOI

Abstract

Synthetic aperture radar (SAR) ocean imagery panoptic segmentation provides a more all-round scene understanding than object detection and instance segmentation. However, it has not been extensively studied yet. We find that the stochastic queries of Transformer-based models from computer vision community lead to poor learning efficiency. Thus, we propose a novel Tri-State prototype self-distillation (TSP-SD) framework that replaces purely random query representations with Tri-State statistical Gaussian prototypes. Specifically, we decouple features into boundary, structure, and context via a Tri-State Gaussian prototype extractor (TGPE) to obtain hier-archical statistical prototypes for progressive evolution. Then, a cascaded statistical query injection (CSQI) decoder utilizes these proto-types to progressively guide the decoding process. Moreover, to reduce training uncertainty, a relative entropy alignment loss (REAL) is proposed to minimize the difference between the teacher and student. Experiments on SPSD and HRSID-PAN show that TSP-SD achieves 86.71% and 83.12% PQ, outperforming Mask2Former by 4.88% and 2.01%, respectively. Incorporating sea-land context further improves ship discrimination in complex coastal scenes.

Topics & Concepts

Computer visionComputer scienceSegmentationArtificial intelligenceSynthetic aperture radarImage segmentationRemote sensingRadar imagingPanopticonFeature extractionObject detectionScale-space segmentationGeologyFeature (linguistics)VisualizationInverse synthetic aperture radarBackscatter (email)Side looking airborne radarImage processingRemote-Sensing Image ClassificationOil Spill Detection and MitigationAdvanced SAR Imaging Techniques