MAH-YOLO: an enhanced YOLOv8n framework for loess landslide detection with multi-attention mechanisms
The Loess Plateau, with its fragile ecological environment and frequent landslides, poses severe risks to both ecological safety and human life. Accurate and efficient landslide detection is essential for disaster prevention and sustainable regional development. This study proposes an enhanced targe...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Taylor & Francis Group
2025-08-01
|
| Series: | International Journal of Digital Earth |
| Subjects: | |
| Online Access: | https://www.tandfonline.com/doi/10.1080/17538947.2025.2536666 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The Loess Plateau, with its fragile ecological environment and frequent landslides, poses severe risks to both ecological safety and human life. Accurate and efficient landslide detection is essential for disaster prevention and sustainable regional development. This study proposes an enhanced target detection architecture, MAH-YOLO, designed for the precise identification of eolian landslides in complex environments. The MAH-YOLO model integrates multi-attention mechanisms (Halo Attention and Global Attention) with a lightweight design (MobileNetv3) to timely optimise landslide detection with high precision. Performance evaluations against Faster-RCNN and YOLOv8n demonstrated substantial improvements in key metrics, including precision (95.68%), recall (80.13%), and mean average precision (mAP) at 88.25%. These metrics were 5.43, 6.01 percentage points, 10.88, 4.96 percentage points, and 5.04, 4.87 percentage points higher than Faster-RCNN and YOLOv8n, respectively. Additionally, MAH-YOLO, a low complexity model with only 3.35M parameters and 16.1G FLOPs was used, balancing efficiency and accuracy. The ablation experiments confirmed the effectiveness of both the multi-attention mechanism and lightweight design. The proposed MAH-YOLO architecture excels at capturing intricate textures and detailed features, offering reliable support for accurate landslide detection. Our work provides valuable insights for intelligent monitoring and early warning systems of geological disasters on the Loess Plateau. |
|---|---|
| ISSN: | 1753-8947 1753-8955 |