COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images
Semantic segmentation of ultra-large remote sensing images faces substantial memory challenges, with virtually no previous work successfully addressing the preservation of global context in such large-scale imagery. Despite the critical importance of maintaining long-range relationships, GPU memory...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-08-01
|
| Series: | International Journal of Applied Earth Observations and Geoinformation |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1569843225003905 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1839604051960922112 |
|---|---|
| author | Trung Dung Nguyen Zhen He Wei Xiang Rajalakshmi Rajasekaran |
| author_facet | Trung Dung Nguyen Zhen He Wei Xiang Rajalakshmi Rajasekaran |
| author_sort | Trung Dung Nguyen |
| collection | DOAJ |
| description | Semantic segmentation of ultra-large remote sensing images faces substantial memory challenges, with virtually no previous work successfully addressing the preservation of global context in such large-scale imagery. Despite the critical importance of maintaining long-range relationships, GPU memory limitations have forced researchers to process these massive images as small disconnected patches. Our novel framework represents a breakthrough, being the first approach (with only one prior incomplete attempt in literature) to efficiently process large 4K patches on a single 24 GB GPU while preserving crucial large-scale image context. Unlike conventional tiling methods, our approach maintains global context through an innovative two-stage process: first extracting features from smaller 1K patches using encoders from powerful segmentation models (SegFormer and the Segment Anything Model), then assembling these embeddings to train a 4K decoder. This strategic separation dramatically reduces memory requirements while maintaining long-range dependencies. Through extensive experiments and ablation studies on the FBP, GID and Inria benchmark datasets, our approach achieves unprecedented performance, significantly outperforming all existing methods in both accuracy and memory efficiency. |
| format | Article |
| id | doaj-art-1a6c0a00bfe34f5abc9d97ea87f7c4a8 |
| institution | Matheson Library |
| issn | 1569-8432 |
| language | English |
| publishDate | 2025-08-01 |
| publisher | Elsevier |
| record_format | Article |
| series | International Journal of Applied Earth Observations and Geoinformation |
| spelling | doaj-art-1a6c0a00bfe34f5abc9d97ea87f7c4a82025-08-02T04:46:40ZengElsevierInternational Journal of Applied Earth Observations and Geoinformation1569-84322025-08-01142104743COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing imagesTrung Dung Nguyen0Zhen He1Wei Xiang2Rajalakshmi Rajasekaran3Corresponding author.; La Trobe University, Department of Computer Science and Information Technology, Melbourne, AustraliaLa Trobe University, Department of Computer Science and Information Technology, Melbourne, AustraliaLa Trobe University, Department of Computer Science and Information Technology, Melbourne, AustraliaLa Trobe University, Department of Computer Science and Information Technology, Melbourne, AustraliaSemantic segmentation of ultra-large remote sensing images faces substantial memory challenges, with virtually no previous work successfully addressing the preservation of global context in such large-scale imagery. Despite the critical importance of maintaining long-range relationships, GPU memory limitations have forced researchers to process these massive images as small disconnected patches. Our novel framework represents a breakthrough, being the first approach (with only one prior incomplete attempt in literature) to efficiently process large 4K patches on a single 24 GB GPU while preserving crucial large-scale image context. Unlike conventional tiling methods, our approach maintains global context through an innovative two-stage process: first extracting features from smaller 1K patches using encoders from powerful segmentation models (SegFormer and the Segment Anything Model), then assembling these embeddings to train a 4K decoder. This strategic separation dramatically reduces memory requirements while maintaining long-range dependencies. Through extensive experiments and ablation studies on the FBP, GID and Inria benchmark datasets, our approach achieves unprecedented performance, significantly outperforming all existing methods in both accuracy and memory efficiency.http://www.sciencedirect.com/science/article/pii/S1569843225003905Semantic segmentationUltra large imagesRemote sensingComputer visionLand coverGlobal context |
| spellingShingle | Trung Dung Nguyen Zhen He Wei Xiang Rajalakshmi Rajasekaran COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images International Journal of Applied Earth Observations and Geoinformation Semantic segmentation Ultra large images Remote sensing Computer vision Land cover Global context |
| title | COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images |
| title_full | COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images |
| title_fullStr | COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images |
| title_full_unstemmed | COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images |
| title_short | COSMOS: Context-preserving satellite memory-optimized segmentation system for ultra large-scale remote sensing images |
| title_sort | cosmos context preserving satellite memory optimized segmentation system for ultra large scale remote sensing images |
| topic | Semantic segmentation Ultra large images Remote sensing Computer vision Land cover Global context |
| url | http://www.sciencedirect.com/science/article/pii/S1569843225003905 |
| work_keys_str_mv | AT trungdungnguyen cosmoscontextpreservingsatellitememoryoptimizedsegmentationsystemforultralargescaleremotesensingimages AT zhenhe cosmoscontextpreservingsatellitememoryoptimizedsegmentationsystemforultralargescaleremotesensingimages AT weixiang cosmoscontextpreservingsatellitememoryoptimizedsegmentationsystemforultralargescaleremotesensingimages AT rajalakshmirajasekaran cosmoscontextpreservingsatellitememoryoptimizedsegmentationsystemforultralargescaleremotesensingimages |