Synthetic aperture radar (SAR) image change detection plays a crucial role in monitoring environmental changes and landform evolution. However, existing methods often struggle to retain details while reducing computational cost, or lack the ability to effectively incorporate global information, especially under complex and noisy conditions. To address these challenges, we propose a novel global robust structured graph approach with difference-aware filtering (GRSG-DAF), which effectively enhances the detection of changed areas by combining pixel-level information with the advantages of the superpixel-based structured graph. First, we construct a structured graph by integrating difference information, utilizing superpixel co-segmentation and adaptive affinity weighting, thus significantly reducing computational complexity and preserving critical structural patterns. Second, an image reconstruction process is implemented, utilizing a feature propagation mechanism to improve the contextual representation of the image. Finally, a difference-aware guided filtering process is developed by integrating the inherent guidance from the original pixel-level image, preserving boundary structures and spatial details. Experimental results demonstrate that our proposed method outperforms state-of-the-art methods on five benchmark datasets, achieving higher accuracy while balancing effectiveness and efficiency in change detection.

Federated learning (FL) enables multiple clients with distributed data sources to collaboratively train a shared model without compromising data privacy. However, existing FL paradigms face challenges due to heterogeneity in client data distributions and system capabilities. Personalized federated learning (pFL) has been proposed to mitigate these problems, but often requires a shared model architecture and a central entity for parameter aggregation, resulting in scalability and communication issues. More recently, model-heterogeneous FL has gained attention due to its ability to support diverse client models, but existing methods are limited by their dependence on a centralized framework, synchronized training, and publicly available datasets. To address these limitations, we introduce Federated Peer-Adaptive Ensemble Learning (FedPAE), a fully decentralized pFL algorithm that supports model heterogeneity and asynchronous learning. Our approach utilizes a peer-to-peer model sharing mechanism and ensemble selection to achieve a more refined balance between local and global information. Experimental results show that FedPAE outperforms existing state-of-the-art pFL algorithms, effectively managing diverse client capabilities and demonstrating robustness against statistical heterogeneity.

Change detection (CD) is a crucial application in the field of remote sensing. Most current CD methods are based on deep learning and revolve around multispectral data. However, a common issue arising from these methods is the large number of model parameters due to the high dimensionality of the data channels. In this paper, we propose a lightweight Siamese network structure that minimally incorporates conventional convolutional layers. Additionally, in terms of data preprocessing, we select the R, G, and B channels of multispectral data for dehazing processing, and employ the processed data as inputs to the network. Experimental results illustrate the outstanding effectiveness and efficiency of the proposed method.

Flood disasters are a major factor threatening agriculture, human life, and property safety. Suppose the areas affected by flood disasters can be effectively delineated and reasonably predicted. In that case, it will not only be beneficial for agricultural production but also provide convenience for disaster prevention and relief. The Segment Anything Model (SAM) emerged, providing innovative ideas for many visual tasks. SAM has excellent feature extraction ability in network models, allowing it to adapt to different scenes and effectively segment various objects, so it also has great application prospects in remote sensing images. Therefore, this article utilizes the excellent feature extraction ability of the SAM to enable the model to adapt to downstream tasks of remote sensing image segmentation. This article will use the decoder structure of CycleGAN as the decoder. Due to the large proportion of background in remote sensing images, this article also enhances the loss function to suit remote sensing image tasks better. The MMFlood dataset in this article consists of Synthetic Aperture Radar (SAR) images combined with a digital elevation model (DEM). The experimental results demonstrate improved performance with the assistance of SAM compared to Unet++.

Change detection (CD) methods for remote sensing images based on deep learning have garnered increasing research attention. However, existing deep learning approaches are often tailored for specific types of sensors. Extending these methods to dual-sensor scenarios presents challenges, including difficulties in data fusion and an increase in parameter numbers. To address these challenges, we propose a novel dual-stream encoder–decoder CD network architecture. In the encoder, the architecture comprises a shared-weight Siamese Unet stream for each sensor, with unique weights for different sensors. Before the decoder, a 3-D attention module (3-D AM) is incorporated, processing encoder outputs and fusing features from different streams. In addition, to mitigate the increased model parameter numbers due to the use of dual sensors, we propose a lightweight Unet architecture along with a time-difference structure in each stream. The proposed model is evaluated across multiple scenarios on a dual-sensor CD dataset, yielding an F1 score of 0.572 and the parameter number of 0.91 M. These results showcase high performance on a cost-effective level. Our code is available athttps://github.com/CodeofHuang/SmaDS_SiamUnet.

We consider a non-convex constrained optimization problem, where the objective function is weakly convex and the constraint function is either convex or weakly convex. To solve this problem, we consider the classical switching subgradient method, which is an intuitive and easily implementable first-order method whose oracle complexity was only known for convex problems. This paper provides the first analysis on the oracle complexity of the switching subgradient method for finding a nearly stationary point of non-convex problems. Our results are derived separately for convex and weakly convex constraints. Compared to existing approaches, especially the double-loop methods, the switching gradient method can be applied to non-smooth problems and achieves the same complexity using only a single loop, which saves the effort on tuning the number of inner iterations.