Research on Landslide Identification in Transmission Corridors Using a YOLOv5-Based Deep Learning Detection Framework

ABSTRACT To address the challenges of intelligent landslide hazard identification along transmission corridors in alpine canyon regions, this study proposes a deep learning detection framework based on the YOLOv5 model. A custom-annotated landslide dataset was developed, complemented by optimised training protocols. Model performance was evaluated using comprehensive metrics, including F1-score curves, precision curves (P-curves), recall curves (R-curves), and precision-recall curves (PR-curves). Taking the Erlang Mountain-Kangding Dadu River transmission corridor as a case study, the framework achieved an accuracy of 85.7%, identifying 28 landslide instances with one false positive and three false negatives, as verified through remote sensing interpretation and field surveys. Results confirm the YOLOv5 model’s efficacy in detecting landslide features within complex geological settings, providing a robust approach for automated monitoring in heterogeneous terrains. This research offers a theoretical and methodological foundation for intelligent geological hazard identification in western China, supporting enhanced disaster mitigation for critical transmission infrastructure.