In smoke interference scenarios, visible image degradation significantly limits the availability of detailed information for fusion. Moreover, a single fusion network often struggles to effectively extract complementary and key information. To overcome these challenges, we propose a infrared and visible image fusion method tailored for smoke interference, termed the Repeated Key Feature Embedding Fusion Network (REFusion). To fully leverage low-frequency and high-frequency information from source images, we design the network with a pre-fusion stage and a fine-fusion stage. To address the heterogeneity between infrared and visible images, we introduce a Cross-Source Context Association Network (CSCAN). By integrating the self-attention mechanism and a cascading pyramid pooling module, we compute attention interaction weights across modes to capture the most relevant context for fusion. To prevent the fusion from favoring one source disproportionately, we enhance the fine-fusion stage with a Differential Information Reinforcement Module (DIRM) and a three-branch network. This structure establishes long-term dependencies between complementary information and excels in capturing high-frequency details. We further design a tailored loss function to enhance the quality of fused images. Extensive experiments demonstrate that REFusion surpasses nine state-of-the-art methods in both visual quality and quantitative metrics, particularly in reproducing scenes obscured by smoke.

Landslides are common and highly threatening geological disasters, and detecting landslide occurrence areas using remote sensing images has important social significance for disaster warning, prediction, and proactive rescue. Currently, the most effective and direct application is in-orbit real-time detection, which obtains the boundary, location and other relevant information of the post-disaster areas by performing in-orbit calculations on the raw images, thereby shortening the relief cycle. However, challenges associated with this application are high efficiency and portability, while overcoming complex surface distributions to ensure high accuracy. In this article, we propose a lightweight network framework and deploy it on a civilian QK series environmental monitoring satellite, transmitting the detection results in graphic form to the ground station. This framework addresses the problems of parameter redundancy and low real-time performance in existing embedded device landslide detection methods. The model adopts a conditional attention mechanism, multifeature fusion, and asymmetric upsampling modules (AUMs). It employs only 3.045 MB of parameters, with a mean intersection over union (MIoU) of 88.38% and an F1-scores of 93.77%, striking a good balance between computational efficiency and accuracy, and its performance can meet the requirements of in-orbit applications. Additionally, we have labeled and published landslide samples that occurred in Luding, Sichuan Province, Southwest China, in September 2022, which is available athttps://github.com/fuadou/project/tree/main/Luding_datasets.

Landslide, a kind of destructive natural disaster, often occurs in the mountainous areas of China. Landslide information instant collection plays an important role in taking appropriate remedial measures and personnel evacuation. In recent years, the use of Convolutional Neural Network (CNN) for landslide regional detection achieved good performance, however, most CNN-based methods had no regard for the internal connection of the cover materials in the disaster occurrence area. Moreover, the information revealed by the internal deformation features was ignored, and the same surface object in the image presents different features under different illumination, environment and resolution, which makes it difficult to extract the structural features of landslide images. In this paper, we propose a novel graph convolutional network for landslide detection, inspired by attention mechanism’s ability to focus on selective information supplemented with both different channels. The global maximum node connection strategy with positive and negative connectivity makes the Graph Convolution Network (GCN) more portable, which is used as the basic unit of graph feature propagation to construct a multi-layer residual connection module. In order to learn interactively and spread graph information, channel dimension is added to make the boundary of features between classes more discriminative. Extensive experiments on Sichuan province and Bijie landslide datasets show that our proposed method outperforms other detection models and achieves high precision and accuracy. In addition, we also carried out landslide detection for Zhaotong of Yunnan Province on GF-2 original images to prove the effectiveness and applicability of the algorithm.

Graph Neural Networks (GNNs), as an effective learning framework for graph structure data representation, have been applied to hyperspectral images (HSIs) classification tasks. Among the variants of GNNs, Graph Attention Networks (GATs) have achieved state-of-the-art node prediction performance by learning to assign dense attention coefficients to all node neighbors for feature aggregation. However, due to the complexity of land distribution and the high dimension of HSIs data, it is difficult to identify different coverage categories by employing GATs directly, and the inadequacy of spectral information also affects the classification accuracy. To address the above application problems, we proposed a multi-stage superpixel-guided sparse GAT for HSI classification (named MSG-SGAT). Specifically, we create the adjacency connection graphs of different stages from the superpixel representation so as to effectively utilize the spatial topology. A sparse GAT (SGAT) module is designed to trim the graph by spectral sparsity to remove some task-irrelevant edges and assign a unique attention coefficient to each remaining edge. Sparse subgraphs are obtained by identifying the noise/task-independent edges and aggregating information. Moreover, SGAT is concatenated with the spectral branch for feature fusion and update, and pixel-level feature refinement is performed at the end of this network. MSG-SGAT mines the features of HSIs from the perspective of multi-scale hierarchy with low calculation cost and high efficiency. A series of comparative experiments on four benchmark datasets demonstrate that the performance of our proposed method is competitive with others.