Haonan Guo

Autonomous feature extraction from high-resolution (HR) Remote Sensing (RS) imagery presents complex spatial–spectral heterogeneity and diverse land-cover patterns, challenging conventional single-architecture models that lack multiscale feature diversity and cross-domain generalization. Convolutional neural networks (CNNs) effectively capture localized hierarchical structures, whereas transformers model long-range contextual dependencies via self-attention. However, using these architectures independently restricts their complementary potential. This study proposes FWDNNet (Fused Weights Deep Neural Tokenization Networks), a fusion framework introducing a CNN-to-token conversion mechanism that transforms convolutional feature maps from heterogeneous encoders into transformer-compatible token sequences. Beyond standard concatenation, FWDNNet integrates a TensorDot-based fusion module to learn high-order cross-architectural interactions, while a shared decoding pathway maintains multiscale feature hierarchies and computational efficiency. Experiments across three diverse datasets show that FWDNNet achieves 58.2 ms inference latency with minimal cross-domain performance degradation (2.3–4.2% mIoU drop). Compared to conventional fusion strategies, FWDNNet attains 21% faster inference and 1.7–2.7% higher mIoU over baseline methods. See Materials at https://github.com/BoazGithub/FWDNNet for reproducibility.

目的显著目标检测是遥感图像理解的重要基础任务。针对像素级标注成本高、视觉基础模型的通用先验难以有效适配复杂遥感场景的问题，提出一种视觉基础模型驱动的低标注遥感显著目标检测方法，以提高无标签样本利用效率和检测精度。方法首先，本文构建了ST-SAM3渐进式自训练框架，该框架利用少量标注样本对SAM3进行轻量级微调，通过迭代提示推理获得无标签样本的细化伪掩码，并构建多维度的伪标签可靠性评估策略，从而筛选高质量无标签样本并生成可靠伪标签。进一步地，本文提出DistillMatch半监督蒸馏学习框架。该框架在尺度扰动和旋转扰动下约束学生预测与教师伪标签保持几何一致；同时设计统计信息蒸馏策略，从特征分布层面增强学生模型对遥感显著目标结构和语义响应的建模能力。结果本文在ORSSD、EORSSD、ORSI-4199数据集上，以5%和10%的低标注比例进行实验。结果表明，本文方法在定量指标上显著优于现有半监督方法，验证了渐进式可靠伪标签构建、几何一致性学习与统计信息蒸馏的有效性。结论该研究表明，借助SAM3基础模型的通用分割先验与提示推理能力，可在无需额外人工干预下提升无标签样本的利用效率，缓解深度网络模型对大规模像素级标注的依赖，为低标注空天遥感显著目标检测提供了一种有效可行的技术路径。代码仓库及数据地址：https：//github.com/lyf0801/ST-SAM3.

Although Mamba-based models have demonstrated great potential in hyperspectral image (HSI) classification, existing approaches often rely on patch-wise inputs, causing redundant computation and limiting global spectral–spatial modeling, while the absence of hierarchical representation and explicit interaction further constrains fine-grained fusion. To address these limitations, we propose a Progressive Hybrid Mamba Model (PHDMamba), which is designed to progressively capture long-range spectral-spatial contextual dependencies and gradually fuse spectral and spatial information through adaptive feature interaction, ultimately yielding a unified representation. Specifically, we develop a Progressive Hybrid Mamba Module, which performs stage-wise modeling of long-range dependencies along both spectral and spatial dimensions. In addition, a dedicated Spectral-Spatial Interaction Module is introduced to adaptively integrate contextual spectral and spatial features. Experimental results on three widely used HSI benchmark datasets demonstrate that the proposed method achieves superior classification performance compared to existing approaches. The implementation code will be released publicly.

Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.

Autonomous large-scale high-resolution land cover (HRLC) mapping remains a major challenge in remote sensing due to the scarcity of reliable training data and resolution mismatches between available labels and input of massive of emerging imagery. Existing global land cover products often suffer from coarse spatial resolution and label noise, limiting their utility for fine-scale urban analysis and environmental monitoring. This paper presents C2FNet, a novel Coarse-to-Fine Network designed to generate HRLC maps from noisy, coarse-resolution labels using a weak supervision strategy of the cross-probability. The 2FNet consists of three key modules: (1) Edge Resolution Refinement Backbones (ERRB), which preserve spatial detail via multiscale feature extraction through parallel convolutional branches; (2) Unsupervised Dynamic Shuffle and Diagonal Annotation (UDSDA), which enhances training reliability by identifying confident regions through spatial-consistency analysis and confidence estimation; and (3) a Contrasting Self-Supervised Loss (C2F-Loss) that integrates cross-entropy and cosine similarity terms to mitigate supervision noise and resolution gaps. Evaluations of three benchmark data sets that encompass diverse urban and rural landscapes show that C2FNet achieves state of the art performance, with 80. 01% overall accuracy and a Cohen’s kappa score of 0.7567, outperforming state-of-the-art (SoA) models with weak supervision. Dataset and code: Available here.

Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.