Answers are grounded in this paper and cite their sources.
Landslides occur frequently in plateau and mountainous regions. Remote sensing images acquired over complex terrain typically suffer from strong background interference, weak features of small-scale landslide targets, imbalanced sample distributions, and difficulties in target localization, which make it difficult for existing detection models to balance detection accuracy with lightweight deployment requirements. To address these issues, this paper proposes CSFS-YOLOv8n, a lightweight model with high detection accuracy for remote sensing landslide detection based on the lightweight YOLOv8n network. First, the SCSA attention mechanism is introduced to enhance the network’s ability to perceive weak landslide features and effectively suppress background noise interference in complex terrain. Second, a synergistic optimization of Focal Loss and SIoU loss is adopted to alleviate sample class imbalance while improving the boundary localization accuracy of landslide targets. Furthermore, the lightweight SCConv convolutional module is embedded to strengthen multi-scale feature fusion, and the AdamW optimization strategy is used to further improve convergence stability and generalization capability. Experimental results show that compared with the original YOLOv8n model, CSFS-YOLOv8n improves precision, recall, mAP@0.5, and mAP@0.5:0.95 by 2.2%, 4.5%, 3.7%, and 3.9%, respectively. Meanwhile, the parameter count increases by only 0.14% and the computational cost by only 2.47%, achieving a significant improvement in detection accuracy while maintaining real-time inference capability. The results indicate that the proposed model has strong anti-interference ability and scene adaptability in complex plateau and mountainous remote sensing scenarios, and can provide efficient and reliable technical support for intelligent landslide disaster identification, hazard investigation, and disaster prevention and mitigation.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: