Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models

High-resolution remote sensing imagery plays an essential role in urban management and environmental monitoring, providing detailed insights for applications ranging from land cover mapping to disaster response. Semantic segmentation methods are among the most effective techniques for comprehensive...

Full description

Saved in:
Bibliographic Details
Main Authors: Yiyun Luo, Jinnian Wang, Jean Sequeira, Xiankun Yang, Dakang Wang, Jiabin Liu, Grekou Yao, Sébastien Mavromatis
Format: Article
Language:English
Published: MDPI AG 2025-07-01
Series:Remote Sensing
Subjects:
Online Access:https://www.mdpi.com/2072-4292/17/15/2579
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849765941990129664
author Yiyun Luo
Jinnian Wang
Jean Sequeira
Xiankun Yang
Dakang Wang
Jiabin Liu
Grekou Yao
Sébastien Mavromatis
author_facet Yiyun Luo
Jinnian Wang
Jean Sequeira
Xiankun Yang
Dakang Wang
Jiabin Liu
Grekou Yao
Sébastien Mavromatis
author_sort Yiyun Luo
collection DOAJ
description High-resolution remote sensing imagery plays an essential role in urban management and environmental monitoring, providing detailed insights for applications ranging from land cover mapping to disaster response. Semantic segmentation methods are among the most effective techniques for comprehensive land cover mapping, and they commonly employ ImageNet-based pre-training semantics. However, traditional fine-tuning processes exhibit poor transferability across different downstream tasks and require large amounts of labeled data. To address these challenges, we introduce Denoising Diffusion Probabilistic Models (DDPMs) as a generative pre-training approach for semantic features extraction in remote sensing imagery. We pre-trained a DDPM on extensive unlabeled imagery, obtaining features at multiple noise levels and resolutions. In order to integrate and optimize these features efficiently, we designed a multi-layer perceptron module with residual connections. It performs channel-wise optimization to suppress feature redundancy and refine representations. Additionally, we froze the feature extractor during fine-tuning. This strategy significantly reduces computational consumption and facilitates fast transfer and deployment across various interpretation tasks on homogeneous imagery. Our comprehensive evaluation on the sparsely labeled dataset MiniFrance-S and the fully labeled Gaofen Image Dataset achieved mean intersection over union scores of 42.7% and 66.5%, respectively, outperforming previous works. This demonstrates that our approach effectively reduces reliance on labeled imagery and increases transferability to downstream remote sensing tasks.
format Article
id doaj-art-4a2268a9068a47fbbc091f02bb5b742e
institution DOAJ
issn 2072-4292
language English
publishDate 2025-07-01
publisher MDPI AG
record_format Article
series Remote Sensing
spelling doaj-art-4a2268a9068a47fbbc091f02bb5b742e2025-08-20T03:04:43ZengMDPI AGRemote Sensing2072-42922025-07-011715257910.3390/rs17152579Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion ModelsYiyun Luo0Jinnian Wang1Jean Sequeira2Xiankun Yang3Dakang Wang4Jiabin Liu5Grekou Yao6Sébastien Mavromatis7G-Mod, LIS, CNRS, Aix-Marseille University, 13009 Marseille, FranceInstitute of Aerospace Remote Sensing Innovations, Guangzhou University, Guangzhou 510006, China2ik Company, 13360 Marseille, FranceInstitute of Aerospace Remote Sensing Innovations, Guangzhou University, Guangzhou 510006, ChinaInstitute of Aerospace Remote Sensing Innovations, Guangzhou University, Guangzhou 510006, ChinaInstitute of Aerospace Remote Sensing Innovations, Guangzhou University, Guangzhou 510006, ChinaG-Mod, LIS, CNRS, Aix-Marseille University, 13009 Marseille, FranceG-Mod, LIS, CNRS, Aix-Marseille University, 13009 Marseille, FranceHigh-resolution remote sensing imagery plays an essential role in urban management and environmental monitoring, providing detailed insights for applications ranging from land cover mapping to disaster response. Semantic segmentation methods are among the most effective techniques for comprehensive land cover mapping, and they commonly employ ImageNet-based pre-training semantics. However, traditional fine-tuning processes exhibit poor transferability across different downstream tasks and require large amounts of labeled data. To address these challenges, we introduce Denoising Diffusion Probabilistic Models (DDPMs) as a generative pre-training approach for semantic features extraction in remote sensing imagery. We pre-trained a DDPM on extensive unlabeled imagery, obtaining features at multiple noise levels and resolutions. In order to integrate and optimize these features efficiently, we designed a multi-layer perceptron module with residual connections. It performs channel-wise optimization to suppress feature redundancy and refine representations. Additionally, we froze the feature extractor during fine-tuning. This strategy significantly reduces computational consumption and facilitates fast transfer and deployment across various interpretation tasks on homogeneous imagery. Our comprehensive evaluation on the sparsely labeled dataset MiniFrance-S and the fully labeled Gaofen Image Dataset achieved mean intersection over union scores of 42.7% and 66.5%, respectively, outperforming previous works. This demonstrates that our approach effectively reduces reliance on labeled imagery and increases transferability to downstream remote sensing tasks.https://www.mdpi.com/2072-4292/17/15/2579remote sensing imagery segmentationfine-tuningdenoising diffusion probabilistic modelself-supervised learning
spellingShingle Yiyun Luo
Jinnian Wang
Jean Sequeira
Xiankun Yang
Dakang Wang
Jiabin Liu
Grekou Yao
Sébastien Mavromatis
Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
Remote Sensing
remote sensing imagery segmentation
fine-tuning
denoising diffusion probabilistic model
self-supervised learning
title Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
title_full Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
title_fullStr Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
title_full_unstemmed Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
title_short Label-Efficient Fine-Tuning for Remote Sensing Imagery Segmentation with Diffusion Models
title_sort label efficient fine tuning for remote sensing imagery segmentation with diffusion models
topic remote sensing imagery segmentation
fine-tuning
denoising diffusion probabilistic model
self-supervised learning
url https://www.mdpi.com/2072-4292/17/15/2579
work_keys_str_mv AT yiyunluo labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT jinnianwang labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT jeansequeira labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT xiankunyang labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT dakangwang labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT jiabinliu labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT grekouyao labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels
AT sebastienmavromatis labelefficientfinetuningforremotesensingimagerysegmentationwithdiffusionmodels