Application of the YOLOv11-seg algorithm for AI-based landslide detection and recognition

Area of Science:

• Geosciences
• Artificial Intelligence
• Geological Engineering

Background:

• Landslides are increasing globally, causing significant damage and casualties.
• Accurate landslide detection is crucial for disaster mitigation and prevention.
• Complex terrains and diverse landslide characteristics challenge traditional monitoring methods.

Purpose of the Study:

• To develop an intelligent landslide recognition method using deep learning.
• To enhance landslide boundary detection and pixel-level segmentation accuracy.
• To address challenges posed by complex terrains and varied landslide features.

Main Methods:

• Utilized the YOLOv11-seg deep learning model for landslide recognition.
• Optimized feature extraction and segmentation modules of YOLOv11-seg.
• Applied data augmentation techniques (mirroring, rotation, color adjustment) for improved generalization.

Main Results:

• YOLOv11-seg demonstrated superior performance in detecting complex boundaries and handling occlusion.
• Achieved high F1 scores: 0.8781 for boundary detection and 0.8114 for segmentation.
• Outperformed traditional methods in mean Average Precision (mAP) for bounding box and mask segmentation tasks.

Conclusions:

• The YOLOv11-seg model offers high reliability and adaptability for landslide detection.
• This research provides advanced technological support for intelligent landslide monitoring and risk assessment.
• The method shows significant potential for improving geological disaster monitoring systems.