2026· IEEE Transactions on Geoscience and Remote Sensing· Vol 64, pp. 5631115-5631115· 0 citations· 49 references
Abstract
Vision foundation models (VFMs) pretrained on large-scale datasets have significantly improved performance in remote sensing semantic segmentation. However, existing methods typically rely on full fine-tuning, which requires updating all model parameters. Instead of updating the full parameter set, parameter-efficient fine-tuning (PEFT) achieves competitive performance by optimizing only a small subset of parameters. Despite its success, most existing PEFT methods are mainly designed for natural image tasks and fail to account for the unique multiscale characteristics of remote sensing images. To address these challenges, we propose multi-scale cognitive feature refinement (MsRE) tuning, a novel PEFT method tailored for remote sensing semantic segmentation. In particular, MsRE captures multiscale contextual information by applying cognitive operations with different cognitive fields to intermediate features of the backbone. It then introduces a set of learnable tokens to establish interactions with features at different scales, enabling precise feature refinement and progressive feature propagation across network layers. This mechanism enhances the model’s ability to understand complex remote sensing scenes and improves downstream segmentation performance. With significantly fewer trainable parameters, MsRE provides an efficient yet effective solution for adapting VFMs to remote sensing segmentation tasks. Extensive experiments demonstrate that MsRE achieves competitive segmentation performance with substantially fewer trainable backbone parameters, providing a favorable balance between accuracy and parameter efficiency. The project is available at http://woldier.top/MsRE
Ultrahigh resolution (UHR) remote sensing image segmentation aims to achieve a fine-grained understanding of complex ground scenes. In recent years, vision foundation models (VFMs) have shown strong capability in learning generic structural priors from large-scale visual data, indicating great potential for such fine-grained scene understanding. However, their application to UHR remote sensing images remains limited, as the massive parameter scales of VFMs are difficult to train under UHR remote sensing images. Motivated by the success of the parameter-efficient fine-tuning paradigm on VFMs, we propose a novel parameter-efficient local-context cooperation (PEACE) framework, which significantly reduces trainable parameter overhead while improving segmentation accuracy. In particular, PEACE leverages a shared VFM with minimal trainable parameters to collaboratively process local and corresponding contextual patches partitioned from the UHR remote sensing image. A multireceptive local adapter (MRLA) and a multireceptive context adapter (MRCA) are designed to capture spatial features of local and contextual inputs across multiple receptive fields. Finally, contextual semantics are integrated into local representations. Furthermore, a context-sensitive assistance strategy (CSAS) leverages the correct prediction of the context to effectively overcome the primary limitations of the patch-based training paradigm. Experimental results demonstrate that PEACE effectively exploits VFMs and remote sensing foundation models (RSFMs) with minimal parameter increments and achieves versatility and superior performance across several UHR remote sensing image benchmarks.
Qi Li, Chun-Ju Chen, Jiaxin Cai et al.· IEEE Transactions on Geoscie...· 0 citations
Remote sensing image semantic segmentation (RSISS) has attracted significant attention due to the growing demand for fine-grained land cover information. The Segment Anything Model (SAM), proposed as a foundation vision model, offers strong segmentation performance and generalization capabilities for RSISS tasks. However, existing SAM-based approaches face two limitations: (1) Insufficient adaptation of SAM's features to the diverse characteristics of land cover types. (2) Semantic ambiguity at object boundaries, which hinders accurate delineation. To address these limitations, we propose Frequency and Edge-guided SAM (FE-SAM), a scalable and efficient framework for RSISS. Specifically, we introduce a Frequency-Modulated Adapter (FMA) that adaptively decomposes and modulates frequency-domain features based on the input data. It selectively enhances informative high- and low-frequency components corresponding to different land cover types. Furthermore, to improve SAM's ability to capture fine-grained details, we design EGRefiner, which integrates multi-scale edge-enhanced information extracted from the input image. Extensive experiments on three benchmark datasets demonstrate that FE-SAM outperforms state-of-the-art methods. The source codes are available at: https://github.com/oucailab/FE-SAM.
Feng Gao, Zizhe Pan, Haoting Wang et al.· IEEE Transactions on Geoscie...· 0 citations
Highlights What are the main findings? Review of four emerging foundation model paradigms for remote sensing image segmentation—Transformer-based architectures, state space models (Mamba), prompt-driven segmentation (SAM), and self-supervised or multimodal pre-training analyzing trade-offs in global context modeling, computational efficiency, and cross-modal representation. Synthesis of downstream adaptation strategies, including parameter-efficient fine-tuning (LoRA, adapters), prompt engineering, few-shot and zero-shot learning, open-vocabulary segmentation, and domain adaptation, revealing how each strategy addresses the gap between pre-training and remote sensing requirements. What are the implications of the main findings? Identification of fundamental bottlenecks limiting current models, including the tension between representation generality and remote sensing-specific adaptation, multimodal sensor heterogeneity, and insufficiencies in existing evaluation ecosystems and annotation paradigms. A forward-looking research roadmap toward remote-sensing-native pre-training, lightweight edge-deployable architectures, and unified open-world geospatial foundation models, providing guidance for future research and practical deployment. Abstract Remote sensing image segmentation is a foundational task in Earth observation. With the rapid growth of remote sensing datasets in terms of scale, modality diversity, semantic openness, and spatio-temporal complexity, the field is evolving from task-specific supervised learning toward foundation-model paradigms. Recent advances in foundation models—including Transformer-based architectures, Mamba-based state space models (SSMs), prompt-driven frameworks such as the Segment Anything Model (SAM), and self-supervised or multimodal pre-training—have profoundly reshaped the technical landscape of remote sensing image segmentation. This paper reviews recent progress from the perspectives of dataset evolution, model architectures, and downstream adaptation strategies, covering parameter-efficient fine-tuning, prompt engineering, few-shot and zero-shot learning, open-vocabulary segmentation, and domain adaptation. We further analyze core challenges including the tension between representation generality and remote sensing-specific adaptation, multimodal sensor heterogeneity, and the insufficiency of existing evaluation ecosystems. Finally, we discuss future directions toward remote-sensing-native pre-training, lightweight edge deployment, and unified open-world geospatial foundation models.
Ming Deng, Yongyi Chen, Guanghai Ding et al.· Italian National Conference...· 0 citations
Semantic segmentation is vital for remote sensing applications, where accurate feature extraction and representation are essential. Existing encoder–decoder architectures often fail to fully utilize high-dimensional features and recover fine details during decoding. To address this problem, we propose DeepKANSeg, a novel network based on the Kolmogorov–Arnold network (KAN). KAN’s ability to decompose complex functions into univariate transformations enables flexible modeling of intricate data patterns. Our approach introduces two key innovations: a KAN-based deep feature refinement (DFR) module composed of DeepKAN to capture complex spatial and semantic relationships from high-dimensional features, and a global–local KAN (GLKAN) module replacing the traditional multilayer perceptron (MLP) layers with KAN-based linear layers to enhance fine-grained decoding. To evaluate the effectiveness of the proposed method, extensive experiments are conducted on two well-known fine-resolution remote sensing benchmark datasets, namely ISPRS Vaihingen and ISPRS Potsdam. The results demonstrate that the KAN-enhanced segmentation model achieves superior performance in terms of accuracy compared to state-of-the-art methods. Moreover, the univariate decomposition improves interpretability, making it suitable for explainable learning in remote sensing. The source code for this work will be accessible at https://github.com/sstary/SSRS
Ziyao Wang, Yin Hu, Xiaokang Zhang et al.· IEEE Transactions on Geoscie...· 12 citations· ⚡1
High-resolution remote sensing imagery presents unique challenges for efficient visual understanding, including dense object distributions, severe scale variations, strong background redundancy, and complex spatial structures. Existing deep models often rely on deep architectures or computationally intensive global modeling strategies, limiting their deployment on resource-constrained platforms. In this article, we propose an efficient and lightweight backbone network, termed the cooperative feature modeling network (CFMNet), for high-resolution remote sensing image understanding. CFMNet decomposes feature representations into heterogeneous yet complementary subspaces and models them cooperatively within a unified framework. Specifically, it coordinates channel semantics, structure-aware spatial dependencies, local detail enhancement, and global contextual consistency to improve representation efficiency while suppressing redundant computation. Extensive experiments demonstrate the effectiveness and generality of CFMNet. It achieves 96.13%, 95.50%, and 98.10% Top-1 accuracy on NWPU-RESISC45, aerial image dataset (AID), and UC Merced Land Use dataset (UCM), respectively, 79.82% mAP on DOTA-v1.0, 73.02% mAP on DOTA-v1.5, and 90.82% mAP on HRSC2016, as well as 83.8% mIoU on Vaihingen and 53.8% mIoU on LoveDA, while maintaining low parameter count and computational complexity. A scaling-based Pareto analysis on DOTA-v1.0 and LoveDA further shows that CFMNet variants form a favorable efficiency–accuracy frontier compared with representative lightweight backbones. These results indicate that cooperative modeling of heterogeneous features provides an effective and efficient solution for high-resolution remote sensing image understanding. The code will be released at https://github.com/BEIBEIPRINCESS/CFMNet
Jih-Ming Chen, Haonan Guo, Jun Liu et al.· IEEE Transactions on Geoscie...· 0 citations
Limited by the diversity and complexity of real-world scenes, existing remote sensing change detection methods often suffer from insufficient fine-grained semantic understanding and blurred boundaries of change targets. To address these issues, this paper proposes a boundary-constrained remote sensing change detection network based on vision foundation models (BCNet). BCNet employs a differential modeling approach and multi-branch guidance mechanism to design a differential detail enhancement module, amplifying fine-grained semantic information. Through cross-layer feature alignment, stepwise fusion, and edge-sensitive modeling, it constructs a multi-scale edge enhancement module that enhances perception of minute variations and edge details, fully leveraging the universal semantic representation capabilities of the vision foundation model. In addition, an edge feature constraint mechanism is introduced that applies dual guidance and supervision during the feature fusion and output stages. This mechanism achieves refined delineation of change region boundaries and significantly mitigates the issue of boundary blurring. Experimental results on four mainstream datasets, namely LEVIR-CD, WHU-CD, NJDS and MSRS-CD, demonstrate that BCNet outperforms 13 state-of-the-art methods in terms of key metrics including F1 and IoU. Against the best VFM-based baseline, BCNet obtains F1 score gains of 0.21%, 0.71%, 6.33% and 0.63% on the above four datasets. Specifically, the proposed method exhibits superior detection accuracy and edge detail preservation capabilities in complex regions.
Shenbo Liu, Dongxue Zhao, Huang He et al.· Remote Sensing· 0 citations