LAMB-net: long-range adaptive multiscale boundary optimization network for SAR marine oil spill segmentation

Marine oil spill segmentation using synthetic aperture radar (SAR) imagery is hindered by blurred spill boundaries, large variations in target scales, and false positives induced by oceanic look-alike features. Existing segmentation methods struggle to jointly resolve these practical obstacles: conventional convolutional neural networks are limited in capturing long-range global context, pure Transformer models suffer from high computational cost, and pretrained large vision models show poor domain adaptability to SAR data. Meanwhile, existing hybrid networks suffer from heavy architectural overhead and lack task-specific optimization for marine oil spill interpretation. To address these limitations, this paper proposes LAMB-Net, a Long-range Adaptive Multi-scale Boundary Network for SAR marine oil spill segmentation. Its encoder stacks cascaded LASS blocks to capture local texture and global contextual information under linear computational complexity. Within the decoder, a multiscale adaptive fusion module dynamically aggregates multilevel feature representations. Moreover, a boundary enhancement unit is integrated at the prediction head and coupled with a four-term composite loss function to mitigate jagged oil spill contours, fragmented thin oil film regions, and false alarms from interfering sea surface features. Extensive experiments are conducted on the binary SOS dataset and multiclass M4D dataset. On the SOS dataset, the proposed LAMB-Net achieves mIoU values of 85.75% (PALSAR) and 84.41% (Sentinel-1), with a peak F1-score of 90.42%, yielding superior precision and recall compared with competing baselines. For the five-class M4D dataset, it obtains an overall mIoU of 72.47% and an oil spill class mIoU of 64.77%, and substantially improves the discrimination of confusing targets including ships. Ablation experiments validate the effectiveness of each designed module.

Authors

Institutions

Publication Details

Journal
Journal of Applied Remote Sensing
Published
2026-11-16
DOI
https://doi.org/10.1117/1.jrs.20.046510
Primary Topic
Oil Spill Detection and Mitigation
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

LAMB-net: long-range adaptive multiscale boundary optimization network for SAR marine oil spill segmentation

Yan Miao, Chunlei Wu
8 links
Journal of Applied Remote Sensing
Oil Spill Detection and Mitigation
article

LAMB-net: long-range adaptive multiscale boundary optimization network for SAR marine oil spill segmentation

Yan Miao, Chunlei Wu
article en

Abstract

Marine oil spill segmentation using synthetic aperture radar (SAR) imagery is hindered by blurred spill boundaries, large variations in target scales, and false positives induced by oceanic look-alike features. Existing segmentation methods struggle to jointly resolve these practical obstacles: conventional convolutional neural networks are limited in capturing long-range global context, pure Transformer models suffer from high computational cost, and pretrained large vision models show poor domain adaptability to SAR data. Meanwhile, existing hybrid networks suffer from heavy architectural overhead and lack task-specific optimization for marine oil spill interpretation. To address these limitations, this paper proposes LAMB-Net, a Long-range Adaptive Multi-scale Boundary Network for SAR marine oil spill segmentation. Its encoder stacks cascaded LASS blocks to capture local texture and global contextual information under linear computational complexity. Within the decoder, a multiscale adaptive fusion module dynamically aggregates multilevel feature representations. Moreover, a boundary enhancement unit is integrated at the prediction head and coupled with a four-term composite loss function to mitigate jagged oil spill contours, fragmented thin oil film regions, and false alarms from interfering sea surface features. Extensive experiments are conducted on the binary SOS dataset and multiclass M4D dataset. On the SOS dataset, the proposed LAMB-Net achieves mIoU values of 85.75% (PALSAR) and 84.41% (Sentinel-1), with a peak F1-score of 90.42%, yielding superior precision and recall compared with competing baselines. For the five-class M4D dataset, it obtains an overall mIoU of 72.47% and an oil spill class mIoU of 64.77%, and substantially improves the discrimination of confusing targets including ships. Ablation experiments validate the effectiveness of each designed module.

Journal of Applied Remote SensingVol. 20(04)
China University of Petroleum, East China (CN)
Life below water
Oil Spill Detection and Mitigation
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon