Remote sensing image semantic segmentation plays a crucial role in various fields, such as environmental monitoring, urban planning, and agricultural land classification. However, most current research primarily focuses on utilizing the spatial and spectral information of single-temporal remote sensing images, neglecting the valuable temporal information present in historical image sequences. In fact, historical images often contain valuable phenological variations in land features, which exhibit diverse patterns and can significantly benefit from semantic segmentation tasks. This paper introduces a semantic segmentation framework for satellite image time series (SITS) based on dilated convolution and a Transformer encoder. The framework includes spatial encoding and temporal encoding. Spatial encoding, utilizing dilated convolutions exclusively, mitigates the loss of spatial accuracy and the need for up-sampling, while allowing for the extraction of rich multi-scale features through a combination of different dilation rates and dense connections. Temporal encoding leverages a Transformer encoder to extract temporal features for each pixel in the image. To better capture the annual periodic patterns of phenological phenomena in land features, position encoding is calculated based on the image’s acquisition date within the year. To assess the performance of this framework, comparative and ablation experiments were conducted using the PASTIS dataset. The experiments indicate that this framework achieves highly competitive performance with relatively low optimization parameters, resulting in an improvement of 8 percentage points in the mean Intersection over Union (mIoU).
The continuous impact of global climate change and human activities on the natural environment has significantly increased the frequency and complexity of emergencies and natural disasters in urban areas, posing higher requirements for the existing emergency management system. The rise of large language models (LLMs) provides a novel solution for emergency management to become intelligent. This article systematically analyzes the latest application progress of LLMs in disaster emergency management and discusses in detail the key role of LLMs in various stages before, during, and after disasters. In addition, this article puts forward the possible application framework and technical route of LLMs in emergency management in the future. Finally, this article summarizes the current challenges of applying LLMs to intelligent emergency management and outlines valuable research directions for the future.
As a natural event that poses a serious threat to human life, property, and the natural ecology, the effective identification, assessment, and early prevention of geological hazards are crucial. Hunan Province in China is a region with a high incidence of geological hazards, exhibiting complex chain-generated characteristics due to the influence of terraced topography, heavy rainfall, and human activities. Existing landslide monitoring methods have insufficient ability to capture weak deformation at small spatial scales, making it challenging to identify landslide disaster precursors in this region effectively. This paper proposes a multi-resolution SAR collaborative monitoring method using SBAS-InSAR technology for wide-area screening, followed by a joint PS/DS-InSAR processing framework to identify weak deformation signals at small spatial scales. Using 2441 registered geohazard sites in the work area as the background dataset, wide-area InSAR monitoring and remote-sensing interpretation delineated 180 suspected geohazard target areas. Field investigation confirmed 83 of the 180 candidate target zones as active hidden-danger points, corresponding to a field-confirmed rate of 46.11% among the interpreted candidates.