Abstract: [Purposes]Landslide recognition in remote sensing images is crucial in geological disaster detection and emergency response.However,existing methods often face challenges such as limited detection accuracy and poor recognition of small-scale landslide targets.Based on the YOLO11s model,this paper proposes an improved landslide detection algorithm named YOLO11s-RDS,which balances detection accuracy,efficiency,and multi-scale feature extraction capabilities.[Methods]The algorithm enhances the model's detection ability for complex landslide features and small-scale landslide targets by integrating the structural reparameterization module RepVGG into the YOLO11s backbone network,introducing the dynamic upsampler DySample into the neck network,and inserting a multi-scale sequence feature fusion module between the neck network and the head network.[Findings]The experimental results on the remote sensing landslide dataset show that the improved YOLO11s-RDS model achieves improvements of 1.3,8.8,5.5,5.5,and 5.7 percentage points in precision,recall,mAP_0.5,mAP_0.5:0.95,and F1-score,respectively,compared to the original model.[Condusions]Compared to other models in the YOLO series,the improved model YOLO11s-RDS demonstrates strong overall performance in landslide identification in remote sensing imagery.

Key words: YOLO11s model, landslide detection, RepVGG, dynamic upsampler, multi-scale feature fusion