The escalation of climate change, population growth, and urbanisation have amplified the frequency and severity of natural hazards such as landslides. These events cause devastating impacts, such as loss of life, destruction of infrastructure, and environmental damage (e.g., loss of soil, habitat destruction). By leveraging deep learning approaches and their capabilities with remote sensing data, natural hazard events can be effectively detected and monitored. This could provide key information for early warning systems, allowing for better management and assessment of hazard risks. However, detecting and monitoring natural hazards such as landslides remains crucial but challenging. To effectively mitigate landslide risks, we propose a landslide detection network based on the integration of a pure transformer network with a Siamese U-shaped structure and an intermediate multilevel fusion strategy, called DETSlideNet (Dual-Encoder based Transformer for LandSlide), to detect landslide-affected regions accurately. Our proposed approach achieves state-of-the-art (SOTA) performance by leveraging satellite and Unmanned Aerial Vehicle (UAV) imagery fusion through multilevel fusion, achieving a mean dice score exceeding 90%.
DETSlideNet: Dual-Encoder SwinTransformer for Scalable Landslides Detection and Monitoring
Gamba P.;
2025-01-01
Abstract
The escalation of climate change, population growth, and urbanisation have amplified the frequency and severity of natural hazards such as landslides. These events cause devastating impacts, such as loss of life, destruction of infrastructure, and environmental damage (e.g., loss of soil, habitat destruction). By leveraging deep learning approaches and their capabilities with remote sensing data, natural hazard events can be effectively detected and monitored. This could provide key information for early warning systems, allowing for better management and assessment of hazard risks. However, detecting and monitoring natural hazards such as landslides remains crucial but challenging. To effectively mitigate landslide risks, we propose a landslide detection network based on the integration of a pure transformer network with a Siamese U-shaped structure and an intermediate multilevel fusion strategy, called DETSlideNet (Dual-Encoder based Transformer for LandSlide), to detect landslide-affected regions accurately. Our proposed approach achieves state-of-the-art (SOTA) performance by leveraging satellite and Unmanned Aerial Vehicle (UAV) imagery fusion through multilevel fusion, achieving a mean dice score exceeding 90%.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
