Related Experiment Videos
Dual-stream feature fusion network for landslide detection
Kaleem Ullah1, Xie Tao2,3,4, Tahir Mahmood5
1School of Remote Sensing and Geomatics Engineering, Nanjing University of Information Science and Technology, 210044, Nanjing, Jiangsu, China.
Scientific Reports
|July 18, 2026
Summary
A new Dual-Stream Feature Fusion Network (DSFF-Net) improves landslide detection from remote sensing data. This efficient deep learning model balances accuracy and computational cost for better landslide mapping.
Area of Science:
- Geosciences
- Remote Sensing
- Artificial Intelligence
Background:
- Landslides pose significant threats to infrastructure and human settlements, especially in mountainous areas.
- Deep learning enhances landslide detection from remote sensing imagery but faces challenges with complex terrains and high computational demands.
Purpose of the Study:
- To develop an accurate and efficient deep learning model for landslide segmentation using remote sensing data.
- To address the limitations of existing models in handling complex terrain and computational resources.
Main Methods:
- Proposed a Dual-Stream Feature Fusion Network (DSFF-Net) integrating dual-stream feature extraction, hierarchical feature fusion, and an Enhanced Spatial Attention (ESA) module.
- Evaluated DSFF-Net on the Bijie and Landslide4Sense benchmark datasets.
Main Results:
- DSFF-Net achieved high performance on the Bijie dataset (97.08% F1-score) and competitive results on Landslide4Sense (70.20% F1-score).
- The model demonstrates a strong balance between segmentation accuracy and computational efficiency with approximately 24.51 million parameters.
- Achieved 96.34% precision, 97.83% recall, 97.08% F1-score, and 94.32% mIoU on the Bijie dataset.
Conclusions:
- DSFF-Net offers a reliable approach for automated landslide detection from remote sensing imagery.
- The model's efficiency supports large-scale landslide inventory mapping and environmental monitoring.
- Future research should focus on diverse geographic evaluations and multi-source data integration for enhanced generalization.