Residual Attention UNet with Pyramid Pooling for Landslide Mapping and Comparative Study of UNet Models on Multispectral Data

View at publisher Summarise with VIRAT
Authors: Kathiroli Raja, S. E. R. Sacha Emile R., S. E. R. Sacha Emile R., Kavinkumar N
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Landslide detection in an expedited and accurate way is crucial in disaster management. RAUNeP (Residual Attention U-Net with Pyramid Pooling) is a hybrid deep-learning approach for accurate remote landslide detection and mapping. RAUNeP is a U-Net architecture with residual blocks integrated with attention mechanism and pyramid pooling in the pooling layer. The dataset used for training is multi-spectral and terrain composite data from Landslide4Sense dataset. The results indicate that the proposed RAUNeP significantly improves the detection of areas affected by landslides by reducing false negatives. This improvement is attributed to the ability of the attention mechanism to focus on landslide sites and the improved global-local feature extraction by the pyramid pooling along with the focal Tversky loss. While the U-Net and the Residual Attention U-Net model with attention nodes scored the F1 score of 0.72 and 0.74 and the recall of 0.70 and 0.75, the proposed RAUNeP scores a recall and F1-score of 0.808 and 0.75 respectively. We also infer the inverse relation between vegetation (NDVI) and the occurrence of landslides

Fields of Study & Subjects

Landslides and related hazards Remote-Sensing Image Classification Remote Sensing and Land Use

Publication details

Journal / source
International Conference on Computing, Communication, Control and Cyber-Physical Systems (I5CPS)
Pages
1–6
Publication date
26 February 2026
Language
EN