

This repository contains implementations details about my publications concerning
Advancing Global Landslide Segmentation: A Coupled Multispectral Attention and Data Augmentation Approach Using the novel MRGSLD Dataset.
-A novel global large-scale dataset MRGSLD containing 21 distinct regions (in some data in Africa and Europe regions ) distributed around the globe is created.
- An innovative Multiple Fusion Synthetic Minority Oversampling technique for landslide data augmentation is advised to address the dataset imbalance.
-The MSFAM-ResAttUnet network for landslide segmentation is proposed.
-The feature attention module (FAM) was proposed to avoid the loss of landslide informations during the downsampling and helps the model ignore noisy pixels. While the Multi-Spectral (MS) branch was added to the network to leverage spectral features contained in the remote sensing images (RSI)