Lights-Transformer: An Efficient Transformer-Based Landslide Detection Model for High-Resolution Remote Sensing...

Area of Science:

• Earth Science
• Computer Science
• Artificial Intelligence

Background:

• Remote sensing is crucial for natural disaster management, especially for early landslide detection.
• Deep learning models enhance landslide detection efficiency but struggle with feature extraction and computational complexity.
• Existing models often lose contextual information and require high computational resources.

Purpose of the Study:

• To propose an innovative landslide detection model, Lights-Transformer, to enhance accuracy and efficiency.
• To address challenges of incomplete feature extraction, information loss, and high computational complexity in current models.
• To improve the automated analysis and rapid identification of landslide-affected regions.

Main Methods:

• Developed an encoder-decoder architecture incorporating multi-scale contextual information and an efficient attention mechanism.
• Introduced a Fusion Block for enhanced multi-angle feature fusion and a Light Segmentation Head for faster inference.
• Utilized high-resolution remote sensing images for detailed feature extraction.

Main Results:

• Lights-Transformer achieved superior accuracy, precision, and computational efficiency compared to state-of-the-art models.
• Demonstrated exceptional performance on the GDCLD dataset with an mIoU of 85.11%, accuracy of 97.44%, and F1 score of 91.49%.
• Achieved high recall (91.52%) and precision (91.46%), indicating robust landslide identification capabilities.

Conclusions:

• Lights-Transformer effectively extracts detailed features from remote sensing data for accurate landslide detection.
• The model overcomes limitations of existing methods, offering improved accuracy and efficiency in landslide management.
• The proposed model shows significant potential for enhancing natural disaster monitoring and response systems.